devops : remove clblast + LLAMA_CUDA -> GGML_CUDA

ggml-ci
2024-06-26 19:17:26 +03:00
79 changed files with 1773 additions and 2862 deletions
@@ -17,18 +17,19 @@
  rocmPackages,
  vulkan-headers,
  vulkan-loader,
-  curl,
+  clblast,
  useBlas ? builtins.all (x: !x) [
    useCuda
    useMetalKit
+    useOpenCL
    useRocm
    useVulkan
  ] && blas.meta.available,
  useCuda ? config.cudaSupport,
-  useMetalKit ? stdenv.isAarch64 && stdenv.isDarwin,
+  useMetalKit ? stdenv.isAarch64 && stdenv.isDarwin && !useOpenCL,
  useMpi ? false, # Increases the runtime closure size by ~700M
+  useOpenCL ? false,
  useRocm ? config.rocmSupport,
-  enableCurl ? true,
  useVulkan ? false,
  llamaVersion ? "0.0.0", # Arbitrary version, substituted by the flake

@@ -55,6 +56,7 @@ let
    ++ lib.optionals useCuda [ "CUDA" ]
    ++ lib.optionals useMetalKit [ "MetalKit" ]
    ++ lib.optionals useMpi [ "MPI" ]
+    ++ lib.optionals useOpenCL [ "OpenCL" ]
    ++ lib.optionals useRocm [ "ROCm" ]
    ++ lib.optionals useVulkan [ "Vulkan" ];

@@ -196,19 +198,19 @@ effectiveStdenv.mkDerivation (
      optionals effectiveStdenv.isDarwin darwinBuildInputs
      ++ optionals useCuda cudaBuildInputs
      ++ optionals useMpi [ mpi ]
+      ++ optionals useOpenCL [ clblast ]
      ++ optionals useRocm rocmBuildInputs
      ++ optionals useBlas [ blas ]
-      ++ optionals useVulkan vulkanBuildInputs
-      ++ optionals enableCurl [ curl ];
+      ++ optionals useVulkan vulkanBuildInputs;

    cmakeFlags =
      [
        (cmakeBool "LLAMA_BUILD_SERVER" true)
        (cmakeBool "BUILD_SHARED_LIBS" (!enableStatic))
        (cmakeBool "CMAKE_SKIP_BUILD_RPATH" true)
-        (cmakeBool "LLAMA_CURL" enableCurl)
        (cmakeBool "GGML_NATIVE" false)
        (cmakeBool "GGML_BLAS" useBlas)
+        (cmakeBool "GGML_CLBLAST" useOpenCL)
        (cmakeBool "GGML_CUDA" useCuda)
        (cmakeBool "GGML_HIPBLAS" useRocm)
        (cmakeBool "GGML_METAL" useMetalKit)
@@ -252,6 +254,7 @@ effectiveStdenv.mkDerivation (
        useCuda
        useMetalKit
        useMpi
+        useOpenCL
        useRocm
        useVulkan
        ;
@@ -278,7 +281,7 @@ effectiveStdenv.mkDerivation (
      # Configurations we don't want even the CI to evaluate. Results in the
      # "unsupported platform" messages. This is mostly a no-op, because
      # cudaPackages would've refused to evaluate anyway.
-      badPlatforms = optionals useCuda lib.platforms.darwin;
+      badPlatforms = optionals (useCuda || useOpenCL) lib.platforms.darwin;

      # Configurations that are known to result in build failures. Can be
      # overridden by importing Nixpkgs with `allowBroken = true`.
@@ -9,3 +9,5 @@ contact_links:
  - name: Want to contribute?
    url: https://github.com/ggerganov/llama.cpp/wiki/contribute
    about: Head to the contribution guide page of the wiki for areas you can help with
+
+
@@ -10,10 +10,10 @@ on:
  push:
    branches:
      - master
-    paths: ['.github/workflows/build.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m', '**/*.metal']
+    paths: ['.github/workflows/**', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m']
  pull_request:
    types: [opened, synchronize, reopened]
-    paths: ['.github/workflows/build.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m', '**/*.metal']
+    paths: ['.github/workflows/build.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m']

 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
@@ -47,7 +47,7 @@ jobs:
          sysctl -a
          mkdir build
          cd build
-          cmake -DLLAMA_FATAL_WARNINGS=ON -DGGML_METAL_EMBED_LIBRARY=ON -DLLAMA_CURL=ON -DBUILD_SHARED_LIBS=OFF ..
+          cmake -DLLAMA_FATAL_WARNINGS=ON -DGGML_METAL_EMBED_LIBRARY=ON -DLLAMA_CURL=ON ..
          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)

      - name: Test
@@ -105,7 +105,7 @@ jobs:
          sysctl -a
          # Metal is disabled due to intermittent failures with Github runners not having a GPU:
          # https://github.com/ggerganov/llama.cpp/actions/runs/8635935781/job/23674807267#step:5:2313
-          cmake -B build -DLLAMA_FATAL_WARNINGS=ON -DGGML_METAL=OFF -DLLAMA_CURL=ON -DBUILD_SHARED_LIBS=OFF
+          cmake -B build -DLLAMA_FATAL_WARNINGS=ON -DGGML_METAL=OFF -DLLAMA_CURL=ON
          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)

      - name: Test
@@ -222,7 +222,7 @@ jobs:
        run: |
          mkdir build
          cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_CURL=ON -DBUILD_SHARED_LIBS=OFF
+          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_CURL=ON
          cmake --build . --config Release -j $(nproc)

      - name: Test
@@ -799,7 +799,6 @@ jobs:
          7z x "-o${env:RUNNER_TEMP}" $env:RUNNER_TEMP/sde.tar
          $sde = $(join-path $env:RUNNER_TEMP sde-external-${env:SDE_VERSION}-win/sde.exe)
          cd build
-          $env:LLAMA_SKIP_TESTS_SLOW_ON_EMULATOR = 1
          & $sde -future -- ctest -L main -C Release --verbose --timeout 900

      - name: Determine tag name
@@ -14,7 +14,6 @@ on:
  push:
    branches:
      - master
-    paths: ['.github/workflows/docker.yml', '.devops/*.Dockerfile', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m', '**/*.metal']

 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
@@ -1,9 +1,8 @@
-# date: Wed Jun 26 19:36:34 EEST 2024
+# date: Tue Apr  9 09:17:14 EEST 2024
 # this file is auto-generated by scripts/gen-authors.sh

 0cc4m <picard12@live.de>
 0xspringtime <110655352+0xspringtime@users.noreply.github.com>
-20kdc <asdd2808@gmail.com>
 2f38b454 <dxf@protonmail.com>
 3ooabkhxtn <31479382+3ooabkhxtn@users.noreply.github.com>
 44670 <44670@users.noreply.github.com>
@@ -12,18 +11,14 @@ AT <manyoso@users.noreply.github.com>
 Aarni Koskela <akx@iki.fi>
 Aaron Miller <apage43@ninjawhale.com>
 Aaryaman Vasishta <aaryaman.vasishta@amd.com>
-Abheek Gulati <abheekg@hotmail.com>
 Abhilash Majumder <30946547+abhilash1910@users.noreply.github.com>
 Abhishek Gopinath K <31348521+overtunned@users.noreply.github.com>
 Adithya Balaji <adithya.b94@gmail.com>
 AdithyanI <adithyan.i4internet@gmail.com>
 Adrian <smith.adriane@gmail.com>
 Adrian Hesketh <a-h@users.noreply.github.com>
-Ahmet Zeer <ahmed.zeer@std.yildiz.edu.tr>
 AidanBeltonS <87009434+AidanBeltonS@users.noreply.github.com>
 Aisuko <urakiny@gmail.com>
-Akarshan Biswas <akarshanbiswas@fedoraproject.org>
-Albert Jin <albert.jin@gmail.com>
 Alberto <57916483+albbus-stack@users.noreply.github.com>
 Alex <awhill19@icloud.com>
 Alex Azarov <alex@azarov.by>
@@ -40,24 +35,19 @@ Ali Nehzat <ali.nehzat@thanks.dev>
 Ali Tariq <ali.tariq@10xengineers.ai>
 Alon <alonfaraj@gmail.com>
 AlpinDale <52078762+AlpinDale@users.noreply.github.com>
-Amir <amir_zia@outlook.com>
 AmirAli Mirian <37371367+amiralimi@users.noreply.github.com>
 Ananta Bastola <anantarajbastola@gmail.com>
 Anas Ahouzi <112881240+aahouzi@users.noreply.github.com>
 András Salamon <ott2@users.noreply.github.com>
 Andrei <abetlen@gmail.com>
 Andrew Canis <andrew.canis@gmail.com>
-Andrew Downing <andrew2085@gmail.com>
 Andrew Duffy <a10y@users.noreply.github.com>
 Andrew Godfrey <AndrewGodfrey@users.noreply.github.com>
-Andy Tai <andy-tai@users.noreply.github.com>
 Arik Poznanski <arikpoz@users.noreply.github.com>
 Artem <guinmoon@gmail.com>
-Artem Zinnatullin <ceo@abstractny.gay>
 Artyom Lebedev <vagran.ast@gmail.com>
 Asbjørn Olling <asbjornolling@gmail.com>
 Ásgeir Bjarni Ingvarsson <asgeir@fundinn.org>
-Ashish <1856117+ashishdatta@users.noreply.github.com>
 Ashok Gelal <401055+ashokgelal@users.noreply.github.com>
 Ashraful Islam <ashraful.meche@gmail.com>
 Atsushi Tatsuma <yoshoku@outlook.com>
@@ -67,46 +57,35 @@ BADR <contact@pythops.com>
 Bach Le <bach@bullno1.com>
 Bailey Chittle <39804642+bachittle@users.noreply.github.com>
 BarfingLemurs <128182951+BarfingLemurs@users.noreply.github.com>
-Bartowski <ckealty1182@gmail.com>
 Behnam M <58621210+ibehnam@users.noreply.github.com>
-Ben Ashbaugh <ben.ashbaugh@intel.com>
 Ben Garney <bengarney@users.noreply.github.com>
 Ben Siraphob <bensiraphob@gmail.com>
 Ben Williams <ben@719ben.com>
-Benjamin Findley <39356821+Kartoffelsaft@users.noreply.github.com>
 Benjamin Lecaillon <84293038+blecaillon@users.noreply.github.com>
 Bernat Vadell <hounter.caza@gmail.com>
-Bingan <70050083+binganao@users.noreply.github.com>
 Bodo Graumann <mail@bodograumann.de>
 Bono Lv <lvscar@users.noreply.github.com>
 Borislav Stanimirov <b.stanimirov@abv.bg>
 Branden Butler <bwtbutler@hotmail.com>
 Brian <mofosyne@gmail.com>
 Bruce MacDonald <brucewmacdonald@gmail.com>
-Bryan Honof <bryanhonof@gmail.com>
 CJ Pais <cj@cjpais.com>
 CRD716 <crd716@gmail.com>
-Calvin Laurenson <calvin@laurenson.dev>
 Cameron <csteele@steelecameron.com>
 Cameron Kaiser <classilla@users.noreply.github.com>
-Carolinabanana <140120812+Carolinabanana@users.noreply.github.com>
 Casey Primozic <casey@cprimozic.net>
 Casey Primozic <me@ameo.link>
 CausalLM <148736309+CausalLM@users.noreply.github.com>
 Cebtenzzre <cebtenzzre@gmail.com>
 Chad Brewbaker <crb002@gmail.com>
-Chao Jiang <jc19chaoj@zoho.com>
 Cheng Shao <terrorjack@type.dance>
-Chris Elrod <elrodc@gmail.com>
 Chris Kuehl <ckuehl@ckuehl.me>
 Christian Demsar <christian@github.email.demsar.us>
 Christian Demsar <crasm@git.vczf.us>
 Christian Falch <875252+chrfalch@users.noreply.github.com>
 Christian Kögler <ck3d@gmx.de>
-Christian Zhou-Zheng <59622928+christianazinn@users.noreply.github.com>
 Clark Saben <76020733+csaben@users.noreply.github.com>
 Clint Herron <hanclinto@gmail.com>
-CrispStrobe <154636388+CrispStrobe@users.noreply.github.com>
 Cuong Trinh Manh <nguoithichkhampha@gmail.com>
 DAN™ <dranger003@gmail.com>
 Damian Stewart <d@damianstewart.com>
@@ -116,12 +95,8 @@ Daniel Bevenius <daniel.bevenius@gmail.com>
 Daniel Drake <drake@endlessos.org>
 Daniel Hiltgen <dhiltgen@users.noreply.github.com>
 Daniel Illescas Romero <illescas.daniel@protonmail.com>
-Daniele <57776841+daniandtheweb@users.noreply.github.com>
 DannyDaemonic <DannyDaemonic@gmail.com>
 Dat Quoc Nguyen <2412555+datquocnguyen@users.noreply.github.com>
-Dave <dave-fl@users.noreply.github.com>
-Dave Airlie <airlied@gmail.com>
-Dave Airlie <airlied@redhat.com>
 Dave Della Costa <ddellacosta+github@gmail.com>
 David Friehs <david@friehs.info>
 David Kennedy <dakennedyd@gmail.com>
@@ -129,13 +104,10 @@ David Pflug <david@pflug.email>
 David Renshaw <dwrenshaw@gmail.com>
 David Sommers <12738+databyte@users.noreply.github.com>
 David Yang <davidyang6us@gmail.com>
-Dawid Potocki <github@dawidpotocki.com>
 Dawid Wysocki <62249621+TortillaZHawaii@users.noreply.github.com>
 Dean <Dean.Sinaean@gmail.com>
 Deins <deinsegle@gmail.com>
-Deven Mistry <31466137+deven367@users.noreply.github.com>
 Didzis Gosko <didzis@users.noreply.github.com>
-Djip007 <djip.perois@free.fr>
 Don Mahurin <dmahurin@users.noreply.github.com>
 DooWoong Lee (David) <manics99@naver.com>
 Doomsdayrs <38189170+Doomsdayrs@users.noreply.github.com>
@@ -144,11 +116,8 @@ Dr. Tom Murphy VII Ph.D <499244+tom7@users.noreply.github.com>
 Ebey Abraham <ebey97@gmail.com>
 Ed Lee <edilee@mozilla.com>
 Ed Lepedus <ed.lepedus@googlemail.com>
-Eddie-Wang <wangjinheng1120@163.com>
 Edward Taylor <edeetee@gmail.com>
-Elaine <elaine.zosa@gmail.com>
 Elbios <141279586+Elbios@users.noreply.github.com>
-Elton Kola <eltonkola@gmail.com>
 Engininja2 <139037756+Engininja2@users.noreply.github.com>
 Equim <sayaka@ekyu.moe>
 Eric Sommerlade <es0m@users.noreply.github.com>
@@ -174,47 +143,37 @@ Firat <firatkiral@gmail.com>
 Folko-Ven <71110216+Folko-Ven@users.noreply.github.com>
 Foul-Tarnished <107711110+Foul-Tarnished@users.noreply.github.com>
 Francisco Melo <43780565+francis2tm@users.noreply.github.com>
-Frank Mai <thxcode0824@gmail.com>
 FrankHB <frankhb1989@gmail.com>
-Fred Douglas <43351173+fredlas@users.noreply.github.com>
 Frederik Vogel <Schaltfehler@users.noreply.github.com>
 Gabe Goodhart <gabe.l.hart@gmail.com>
 GainLee <perfecter.gen@gmail.com>
 Galunid <karolek1231456@gmail.com>
 Gary Linscott <glinscott@gmail.com>
 Gary Mulder <gjmulder@gmail.com>
-Gavin Zhao <gavinzhaojw@protonmail.com>
 Genkagaku.GPT <hlhr202@163.com>
 Georgi Gerganov <ggerganov@gmail.com>
 Gilad S <giladgd@users.noreply.github.com>
-Giuseppe Scrivano <giuseppe@scrivano.org>
 GiviMAD <GiviMAD@users.noreply.github.com>
 Govlzkoy <gotope@users.noreply.github.com>
 Guillaume "Vermeille" Sanchez <Guillaume.V.Sanchez@gmail.com>
 Guillaume Wenzek <gwenzek@users.noreply.github.com>
 Guoteng <32697156+SolenoidWGT@users.noreply.github.com>
 Gustavo Rocha Dias <91472747+gustrd@users.noreply.github.com>
-Haggai Nuchi <h.nuchi@gmail.com>
 Halalaluyafail3 <55773281+Halalaluyafail3@users.noreply.github.com>
-Hamdoud Hakem <90524568+hamdoudhakem@users.noreply.github.com>
-HanishKVC <hanishkvc@gmail.com>
 Haohui Mai <ricetons@gmail.com>
 Haoxiang Fei <tonyfettes@tonyfettes.com>
 Harald Fernengel <harald.fernengel@here.com>
 Hatsune Miku <129688334+at8u@users.noreply.github.com>
-HatsuneMikuUwU33 <173229399+HatsuneMikuUwU33@users.noreply.github.com>
 Henk Poley <HenkPoley@gmail.com>
 Henri Vasserman <henv@hot.ee>
 Henrik Forstén <henrik.forsten@gmail.com>
 Herman Semenov <GermanAizek@yandex.ru>
 Hesen Peng <hesen.peng@gmail.com>
 Hoang Nguyen <hugo53@users.noreply.github.com>
-Hong Bo PENG <penghb@cn.ibm.com>
 Hongyu Ouyang <96765450+casavaca@users.noreply.github.com>
 Howard Su <howard0su@gmail.com>
 Hua Jiang <allenhjiang@outlook.com>
 Huawei Lin <huaweilin.cs@gmail.com>
-Hugo Roussel <hugo.rous@gmail.com>
 Ian Bull <irbull@eclipsesource.com>
 Ian Bull <irbull@gmail.com>
 Ian Scrivener <github@zilogy.asia>
@@ -231,10 +190,8 @@ Ivan Stepanov <ivanstepanovftw@gmail.com>
 JH23X <165871467+JH23X@users.noreply.github.com>
 Jack Mousseau <jmousseau@users.noreply.github.com>
 JackJollimore <130917767+JackJollimore@users.noreply.github.com>
-Jaemin Son <woalsdnd@gmail.com>
 Jag Chadha <jagtesh@gmail.com>
 Jakub N <jakubniemczyk97@gmail.com>
-James A Capozzoli <157492257+jac-jim@users.noreply.github.com>
 James Reynolds <magnusviri@users.noreply.github.com>
 Jan Boon <jan.boon@kaetemi.be>
 Jan Boon <kaetemi@gmail.com>
@@ -248,17 +205,12 @@ Jean-Michaël Celerier <jeanmichael.celerier+github@gmail.com>
 Jed Fox <git@jedfox.com>
 Jeffrey Quesnelle <emozilla@nousresearch.com>
 Jesse Jojo Johnson <williamsaintgeorge@gmail.com>
-Jeximo <jeximo@gmail.com>
 Jhen-Jie Hong <iainst0409@gmail.com>
 Jiahao Li <liplus17@163.com>
 Jian Liao <jianliao@users.noreply.github.com>
 JidongZhang-THU <1119708529@qq.com>
 Jinwoo Jeong <33892306+williamjeong2@users.noreply.github.com>
 Jiří Podivín <66251151+jpodivin@users.noreply.github.com>
-Jiří Sejkora <Sejseloid@gmail.com>
-Joan Fontanals <jfontanalsmartinez@gmail.com>
-Joan Fontanals <joan.fontanals.martinez@jina.ai>
-Johan <JohanAR@users.noreply.github.com>
 Johannes Gäßler <johannesg@5d6.de>
 Johannes Rudolph <johannes.rudolph@gmail.com>
 John <78893154+cmp-nct@users.noreply.github.com>
@@ -269,19 +221,15 @@ Jonas Wunderlich <32615971+jonas-w@users.noreply.github.com>
 Jorge A <161275481+jorgealias@users.noreply.github.com>
 Jose Maldonado <63384398+yukiteruamano@users.noreply.github.com>
 Joseph Stahl <1269177+josephst@users.noreply.github.com>
-Josh Ramer <josh.ramer@icloud.com>
 Joyce <joycebrum@google.com>
 Juan Calderon-Perez <835733+gaby@users.noreply.github.com>
 Judd <foldl@users.noreply.github.com>
 Julius Arkenberg <arki05@users.noreply.github.com>
 Jun Jie <71215065+junnjiee16@users.noreply.github.com>
-Junyang Lin <justinlin930319@hotmail.com>
 Juraj Bednar <juraj@bednar.io>
 Justin Parker <jparkerweb@gmail.com>
 Justin Suess <justin.suess@westpoint.edu>
-Justina Cho <justcho5@gmail.com>
 Justine Tunney <jtunney@gmail.com>
-Justine Tunney <jtunney@mozilla.com>
 Juuso Alasuutari <juuso.alasuutari@gmail.com>
 KASR <karim.asrih@gmail.com>
 Kamil Tomšík <info@tomsik.cz>
@@ -294,7 +242,6 @@ Kawrakow <48489457+ikawrakow@users.noreply.github.com>
 Keiichi Tabata <keiichi.tabata@outlook.com>
 Kenvix ⭐ <kenvixzure@live.com>
 Kerfuffle <44031344+KerfuffleV2@users.noreply.github.com>
-Kevin Gibbons <bakkot@gmail.com>
 Kevin Ji <1146876+kevinji@users.noreply.github.com>
 Kevin Kwok <antimatter15@gmail.com>
 Kevin Lo <kevlo@kevlo.org>
@@ -310,7 +257,6 @@ Laura <Tijntje_7@msn.com>
 Lee <44310445+lx200916@users.noreply.github.com>
 Lee Drake <b.lee.drake@gmail.com>
 Leng Yue <lengyue@lengyue.me>
-Leon Knauer <git@leonknauer.com>
 LeonEricsson <70749762+LeonEricsson@users.noreply.github.com>
 Leonardo Neumann <leonardo@neumann.dev.br>
 Li Tan <tanliboy@gmail.com>
@@ -319,26 +265,20 @@ LoganDark <github@logandark.mozmail.com>
 LostRuins <39025047+LostRuins@users.noreply.github.com>
 Luciano <lucianostrika44@gmail.com>
 Luo Tian <lt@basecity.com>
-Lyle Dean <dean@lyle.dev>
 M. Yusuf Sarıgöz <yusufsarigoz@gmail.com>
 Maarten ter Huurne <maarten@treewalker.org>
 Mack Straight <eiz@users.noreply.github.com>
 Maël Kerbiriou <m431.kerbiriou@gmail.com>
 MaggotHATE <clay1326@gmail.com>
-Manuel <44313466+makuche@users.noreply.github.com>
 Marc Köhlbrugge <subscriptions@marckohlbrugge.com>
 Marco Matthies <71844+marcom@users.noreply.github.com>
 Marcus Dunn <51931484+MarcusDunn@users.noreply.github.com>
 Marian Cepok <marian.cepok@gmail.com>
 Mark Fairbairn <thebaron88@gmail.com>
 Marko Tasic <mtasic85@gmail.com>
-Markus Tavenrath <mtavenrath@users.noreply.github.com>
-Martin Delille <martin@delille.org>
 Martin Krasser <krasserm@googlemail.com>
 Martin Schwaighofer <mschwaig@users.noreply.github.com>
 Marvin Gießing <marvin.giessing@gmail.com>
-Masaya, Kato <62578291+msy-kato@users.noreply.github.com>
-MasterYi1024 <39848311+MasterYi1024@users.noreply.github.com>
 Mateusz Charytoniuk <mateusz.charytoniuk@protonmail.com>
 Matheus C. França <matheus-catarino@hotmail.com>
 Matheus Gabriel Alves Silva <matheusgasource@gmail.com>
@@ -347,11 +287,8 @@ Mathijs de Bruin <mathijs@mathijsfietst.nl>
 Matt Clayton <156335168+mattjcly@users.noreply.github.com>
 Matt Pulver <matt.pulver@heavy.ai>
 Matteo Boschini <12133566+mbosc@users.noreply.github.com>
-Mattheus Chediak <shammcity00@gmail.com>
 Matthew Tejo <matthew.tejo@gmail.com>
 Matvey Soloviev <blackhole89@gmail.com>
-Max Krasnyansky <max.krasnyansky@gmail.com>
-Max Krasnyansky <quic_maxk@quicinc.com>
 Maxime <672982+maximegmd@users.noreply.github.com>
 Maximilian Winter <maximilian.winter.91@gmail.com>
 Meng Zhang <meng@tabbyml.com>
@@ -363,41 +300,32 @@ Michael Kesper <mkesper@schokokeks.org>
 Michael Klimenko <mklimenko29@gmail.com>
 Michael Podvitskiy <podvitskiymichael@gmail.com>
 Michael Potter <NanoTekGuy@Gmail.com>
-Michael de Gans <michael.john.degans@gmail.com>
 Michaël de Vries <vriesdemichael@gmail.com>
 Mihai <mihai.chirculescu@yahoo.com>
 Mike <ytianhui2004@gmail.com>
-Mikko Juola <mikjuo@gmail.com>
 Minsoo Cheong <54794500+mscheong01@users.noreply.github.com>
 Mirko185 <mirkosig@gmail.com>
 Mirror Azure <54669636+MirrorAzure@users.noreply.github.com>
 Miwa / Ensan <63481257+ensan-hcl@users.noreply.github.com>
 Mohammadreza Hendiani <hendiani.mohammadreza@gmail.com>
-Mohammadreza Hendiani <mohammad.r.hendiani@gmail.com>
 Murilo Santana <mvrilo@gmail.com>
 Musab Gultekin <musabgultekin@users.noreply.github.com>
 Nam D. Tran <42194884+namtranase@users.noreply.github.com>
-Nathan Epstein <nate2@umbc.edu>
 NawafAlansari <72708095+NawafAlansari@users.noreply.github.com>
 Nebula <infinitewormhole@gmail.com>
-Neo Zhang <14088817+arthw@users.noreply.github.com>
-Neo Zhang <zhang.jianyu@outlook.com>
 Neo Zhang Jianyu <jianyu.zhang@intel.com>
 Neuman Vong <neuman.vong@gmail.com>
 Nexesenex <124105151+Nexesenex@users.noreply.github.com>
 Niall Coates <1349685+Niall-@users.noreply.github.com>
 Nicolai Weitkemper <kontakt@nicolaiweitkemper.de>
-Nicolás Pérez <nicolas_perez@brown.edu>
 Nigel Bosch <pnigelb@gmail.com>
 Niklas Korz <niklas@niklaskorz.de>
-Nikolas <127742645+nneubacher@users.noreply.github.com>
 Nindaleth <Nindaleth@users.noreply.github.com>
 Oleksandr Nikitin <oleksandr@tvori.info>
 Oleksii Maryshchenko <oleksii.maryshchenko@gmail.com>
 Olivier Chafik <ochafik@users.noreply.github.com>
 Ondřej Čertík <ondrej@certik.us>
 Ouadie EL FAROUKI <ouadie.elfarouki@codeplay.com>
-Patrice Ferlet <metal3d@gmail.com>
 Paul Tsochantaris <ptsochantaris@icloud.com>
 Pavol Rusnak <pavol@rusnak.io>
 Pedro Cuenca <pedro@huggingface.co>
@@ -415,14 +343,9 @@ RJ Adriaansen <adriaansen@eshcc.eur.nl>
 Radoslav Gerganov <rgerganov@gmail.com>
 Radosław Gryta <radek.gryta@gmail.com>
 Rahul Vivek Nair <68507071+RahulVivekNair@users.noreply.github.com>
-Raj Hammeer Singh Hada <hammeerraj@gmail.com>
-Ralph Soika <ralph.soika@imixs.com>
 Rand Xie <randxiexyy29@gmail.com>
 Randall Fitzgerald <randall@dasaku.net>
 Reinforce-II <fate@eastal.com>
-Ren Xuancheng <jklj077@users.noreply.github.com>
-Rene Leonhardt <65483435+reneleonhardt@users.noreply.github.com>
-RhinoDevel <RhinoDevel@users.noreply.github.com>
 Riceball LEE <snowyu.lee@gmail.com>
 Richard Kiss <him@richardkiss.com>
 Richard Roberson <richardr1126@gmail.com>
@@ -450,7 +373,6 @@ Rowan Hart <rowanbhart@gmail.com>
 Rune <43761327+Rune-AI@users.noreply.github.com>
 Ryan Landay <rlanday@gmail.com>
 Ryder Wishart <ryderwishart@gmail.com>
-Ryuei <louixs@users.noreply.github.com>
 Rőczey Barnabás <31726601+An0nie@users.noreply.github.com>
 SakuraUmi <yukinon244@gmail.com>
 Salvador E. Tropea <stropea@inti.gob.ar>
@@ -464,7 +386,6 @@ SebastianApel <13675545+SebastianApel@users.noreply.github.com>
 Senemu <10880819+Senemu@users.noreply.github.com>
 Sergey Alirzaev <zl29ah@gmail.com>
 Sergio López <slp@sinrega.org>
-Sertaç Özercan <852750+sozercan@users.noreply.github.com>
 SeungWon Jeong <65549245+redlion0929@users.noreply.github.com>
 ShadovvBeast <ShadovvBeast@gmail.com>
 Shakhar Dasgupta <shakhardasgupta@gmail.com>
@@ -473,7 +394,6 @@ Shijie <821898965@qq.com>
 Shintarou Okada <kokuzen@gmail.com>
 Shouzheng Liu <61452103+lshzh-ww@users.noreply.github.com>
 Shouzheng Liu <lshzh.hi@gmail.com>
-Shuichi Tsutsumi <shuichi0526@gmail.com>
 Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
 Simon Willison <swillison@gmail.com>
 Siwen Yu <yusiwen@gmail.com>
@@ -485,14 +405,11 @@ Someone <sergei.kozlukov@aalto.fi>
 Someone Serge <sergei.kozlukov@aalto.fi>
 Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>
 Spencer Sutton <spencersutton@users.noreply.github.com>
-Srihari-mcw <96763064+Srihari-mcw@users.noreply.github.com>
 Srinivas Billa <nivibilla@gmail.com>
 Stefan Sydow <stefan@sydow.email>
-Steffen Röcker <sroecker@gmail.com>
 Stephan Walter <stephan@walter.name>
 Stephen Nichols <snichols@users.noreply.github.com>
 Steve Grubb <ausearch.1@gmail.com>
-Steven Prichard <spprichard20@gmail.com>
 Steven Roussey <sroussey@gmail.com>
 Steward Garcia <57494570+FSSRepo@users.noreply.github.com>
 Suaj Carrot <72162667+SuajCarrot@users.noreply.github.com>
@@ -517,19 +434,16 @@ Tom C <tom.corelis@gmail.com>
 Tom Jobbins <784313+TheBloke@users.noreply.github.com>
 Tomas <tom.tomas.36478119@gmail.com>
 Tomáš Pazdiora <tomas.pazdiora@gmail.com>
-Tristan Druyen <tristan@vault81.mozmail.com>
 Tristan Ross <rosscomputerguy@protonmail.com>
 Tungsten842 <886724vf@anonaddy.me>
 Tungsten842 <quantmint@protonmail.com>
 Tushar <ditsuke@protonmail.com>
 UEXTM.com <84163508+uextm@users.noreply.github.com>
-Ulrich Drepper <drepper@gmail.com>
 Uzo Nweke <uzoechi@gmail.com>
 Vaibhav Srivastav <vaibhavs10@gmail.com>
 Val Kharitonov <mail@kharvd.com>
 Valentin Konovalov <valle.ketsujin@gmail.com>
 Valentyn Bezshapkin <61702053+valentynbez@users.noreply.github.com>
-Victor Nogueira <felladrin@gmail.com>
 Victor Z. Peng <ziliangdotme@gmail.com>
 Vlad <spitfireage@gmail.com>
 Vladimir <bogdad@gmail.com>
@@ -541,9 +455,7 @@ Weird Constructor <weirdconstructor@gmail.com>
 Welby Seely <welbyseely@gmail.com>
 Wentai Zhang <rchardx@gmail.com>
 WillCorticesAI <150854901+WillCorticesAI@users.noreply.github.com>
-William Tambellini <william.tambellini@gmail.com>
 Willy Tarreau <w@1wt.eu>
-Wouter <9594229+DifferentialityDevelopment@users.noreply.github.com>
 Wu Jian Ping <wujjpp@hotmail.com>
 Wu Jian Ping <wujp@greatld.com>
 Xiake Sun <xiake.sun@intel.com>
@@ -554,8 +466,6 @@ Xiaoyi Chen <cxychina@gmail.com>
 Xingchen Song(宋星辰) <xingchensong1996@163.com>
 Xuan Son Nguyen <thichthat@gmail.com>
 Yann Follet <131855179+YannFollet@users.noreply.github.com>
-Yaroslav <yaroslav.yashin@me.com>
-Yazan Agha-Schrader <mountaiin@icloud.com>
 Yiming Cui <conandiy@vip.qq.com>
 Yishuo Wang <MeouSker77@outlook.com>
 Yueh-Po Peng <94939112+y10ab1@users.noreply.github.com>
@@ -567,7 +477,6 @@ Zane Shannon <z@zcs.me>
 Zay <95888118+isaiahbjork@users.noreply.github.com>
 Zenix <zenixls2@gmail.com>
 Zhang Peiyuan <a1286225768@gmail.com>
-Zheng.Deng <32841220+dengzheng-cloud@users.noreply.github.com>
 ZhouYuChen <zhouyuchen@naver.com>
 Ziad Ben Hadj-Alouane <zied.benhadjalouane@gmail.com>
 Ziang Wu <97337387+ZiangWu-77@users.noreply.github.com>
@@ -575,18 +484,14 @@ Zsapi <martin1.zsapka@gmail.com>
 a-n-n-a-l-e-e <150648636+a-n-n-a-l-e-e@users.noreply.github.com>
 adel boussaken <netdur@gmail.com>
 afrideva <95653597+afrideva@users.noreply.github.com>
-agray3 <agray3@users.noreply.github.com>
 akawrykow <142945436+akawrykow@users.noreply.github.com>
 alexpinel <93524949+alexpinel@users.noreply.github.com>
 alonfaraj <alonfaraj@gmail.com>
-alwqx <kenan3015@gmail.com>
-amd-lalithnc <lalithnc@amd.com>
 andrijdavid <david@geek.mg>
 anon998 <131767832+anon998@users.noreply.github.com>
 anzz1 <anzz1@live.com>
 apaz <aarpazdera@gmail.com>
 apcameron <37645737+apcameron@users.noreply.github.com>
-arch-btw <57669023+arch-btw@users.noreply.github.com>
 arcrank <arcrank@gmail.com>
 arlo-phoenix <140345165+arlo-phoenix@users.noreply.github.com>
 at8u <129688334+at8u@users.noreply.github.com>
@@ -609,17 +514,13 @@ cocktailpeanut <121128867+cocktailpeanut@users.noreply.github.com>
 coezbek <c.oezbek@gmail.com>
 comex <comexk@gmail.com>
 compilade <113953597+compilade@users.noreply.github.com>
-compilade <git@compilade.net>
-cpumaxx <163466046+cpumaxx@users.noreply.github.com>
 crasm <crasm@git.vczf.net>
 crasm <crasm@git.vczf.us>
 daboe01 <daboe01@googlemail.com>
 david raistrick <keen99@users.noreply.github.com>
-ddh0 <dylanhalladay02@icloud.com>
 ddpasa <112642920+ddpasa@users.noreply.github.com>
 deepdiffuser <112834445+deepdiffuser@users.noreply.github.com>
 divinity76 <divinity76@gmail.com>
-dm4 <sunrisedm4@gmail.com>
 dotpy314 <33351922+dotpy314@users.noreply.github.com>
 drbh <david.richard.holtz@gmail.com>
 ds5t5 <145942675+ds5t5@users.noreply.github.com>
@@ -628,7 +529,6 @@ eastriver <lee@eastriver.dev>
 ebraminio <ebraminio@gmail.com>
 eiery <19350831+eiery@users.noreply.github.com>
 eric8607242 <e0928021388@gmail.com>
-fairydreaming <166155368+fairydreaming@users.noreply.github.com>
 fraxy-v <65565042+fraxy-v@users.noreply.github.com>
 github-actions[bot] <github-actions[bot]@users.noreply.github.com>
 gliptic <gliptic@users.noreply.github.com>
@@ -639,7 +539,6 @@ h-h-h-h <13482553+h-h-h-h@users.noreply.github.com>
 hankcs <cnhankmc@gmail.com>
 hoangmit <hoangmit@users.noreply.github.com>
 hongbo.mo <352280764@qq.com>
-hopkins385 <98618192+hopkins385@users.noreply.github.com>
 howlger <eclipse@voormann.de>
 howlger <github@voormann.de>
 hutli <6594598+hutli@users.noreply.github.com>
@@ -650,22 +549,14 @@ hydai <z54981220@gmail.com>
 iSma <ismail.senhaji@gmail.com>
 iacore <74560659+iacore@users.noreply.github.com>
 igarnier <igarnier@protonmail.com>
-intelmatt <61025942+intelmatt@users.noreply.github.com>
 iohub <rickyang.pro@gmail.com>
 jacobi petrucciani <8117202+jpetrucciani@users.noreply.github.com>
-jaime-m-p <167997752+jaime-m-p@users.noreply.github.com>
 jameswu2014 <545426914@qq.com>
-jiez <373447296@qq.com>
 jneem <joeneeman@gmail.com>
-joecryptotoo <80373433+joecryptotoo@users.noreply.github.com>
 johnson442 <56517414+johnson442@users.noreply.github.com>
-jojorne <jojorne@users.noreply.github.com>
 jon-chuang <9093549+jon-chuang@users.noreply.github.com>
 jp-x-g <jpxg-dev@protonmail.com>
-jukofyork <69222624+jukofyork@users.noreply.github.com>
-junchao-loongson <68935141+junchao-loongson@users.noreply.github.com>
 jwj7140 <32943891+jwj7140@users.noreply.github.com>
-k.h.lai <adrian.k.h.lai@outlook.com>
 kaizau <kaizau@users.noreply.github.com>
 kalomaze <66376113+kalomaze@users.noreply.github.com>
 kang <tpdns9032100@gmail.com>
@@ -684,15 +575,11 @@ ldwang <ftgreat@163.com>
 le.chang <cljs118@126.com>
 leejet <leejet714@gmail.com>
 limitedAtonement <limitedAtonement@users.noreply.github.com>
-liuwei-git <14815172+liuwei-git@users.noreply.github.com>
 lon <114724657+longregen@users.noreply.github.com>
-loonerin <132926317+loonerin@users.noreply.github.com>
-luoyu-intel <yu.luo@intel.com>
 m3ndax <adrian.goessl@outlook.com>
 maddes8cht <55592906+maddes8cht@users.noreply.github.com>
 makomk <makosoft@googlemail.com>
 manikbhandari <mbbhandarimanik2@gmail.com>
-maor-ps <154728172+maor-ps@users.noreply.github.com>
 mdrokz <mohammadmunshi@gmail.com>
 mgroeber9110 <45620825+mgroeber9110@users.noreply.github.com>
 minarchist <minarchist@users.noreply.github.com>
@@ -706,19 +593,15 @@ ngc92 <7938269+ngc92@users.noreply.github.com>
 nhamanasu <45545786+nhamanasu@users.noreply.github.com>
 niansa/tuxifan <anton-sa@web.de>
 niansa/tuxifan <tuxifan@posteo.de>
-nickp27 <nb.porter@gmail.com>
 ningshanwutuobang <ningshanwutuobang@gmail.com>
 nold <Nold360@users.noreply.github.com>
 nopperl <54780682+nopperl@users.noreply.github.com>
 nusu-github <29514220+nusu-github@users.noreply.github.com>
 olexiyb <olexiyb@gmail.com>
-omahs <73983677+omahs@users.noreply.github.com>
 oobabooga <112222186+oobabooga@users.noreply.github.com>
 opparco <parco.opaai@gmail.com>
 ostix360 <55257054+ostix360@users.noreply.github.com>
-pengxin99 <pengxin.yuan@intel.com>
 perserk <perserk@gmail.com>
-pmysl <piotr.myslinski@outlook.com>
 postmasters <namnguyen@google.com>
 pudepiedj <pudepiedj@gmail.com>
 qingfengfenga <41416092+qingfengfenga@users.noreply.github.com>
@@ -731,19 +614,16 @@ rhuddleston <ryan.huddleston@percona.com>
 rimoliga <53384203+rimoliga@users.noreply.github.com>
 runfuture <runfuture@users.noreply.github.com>
 sandyiscool <sandyiscool@gmail.com>
-sasha0552 <admin@sasha0552.org>
 semidark <me@semidark.net>
 sharpHL <132747147+sharpHL@users.noreply.github.com>
 shibe2 <shibe@tuta.io>
 singularity <12184989+singularity-s0@users.noreply.github.com>
 sjinzh <sjinzh@gmail.com>
-sjxx <63994076+ylsdamxssjxxdd@users.noreply.github.com>
 slaren <2141330+slaren@users.noreply.github.com>
 slaren <slarengh@gmail.com>
 snadampal <87143774+snadampal@users.noreply.github.com>
 staviq <staviq@gmail.com>
 stduhpf <stephduh@live.fr>
-strawberrymelonpanda <152940198+strawberrymelonpanda@users.noreply.github.com>
 swittk <switt1995@gmail.com>
 takov751 <40316768+takov751@users.noreply.github.com>
 tarcey <cey.tarik@gmail.com>
@@ -756,16 +636,12 @@ uint256_t <konndennsa@gmail.com>
 uint256_t <maekawatoshiki1017@gmail.com>
 unbounded <haakon@likedan.net>
 valiray <133289098+valiray@users.noreply.github.com>
-vik <vikhyatk@gmail.com>
-viric <viric@viric.name>
 vodkaslime <646329483@qq.com>
 vvhg1 <94630311+vvhg1@users.noreply.github.com>
 vxiiduu <73044267+vxiiduu@users.noreply.github.com>
 wbpxre150 <100937007+wbpxre150@users.noreply.github.com>
 whoreson <139810751+whoreson@users.noreply.github.com>
-woachk <24752637+woachk@users.noreply.github.com>
 wonjun Jang <strutive07@gmail.com>
-woodx <124784234+woodx9@users.noreply.github.com>
 wzy <32936898+Freed-Wu@users.noreply.github.com>
 xaedes <xaedes@gmail.com>
 xaedes <xaedes@googlemail.com>
@@ -773,10 +649,7 @@ xloem <0xloem@gmail.com>
 yangli2 <yangli2@gmail.com>
 yuiseki <yuiseki@gmail.com>
 zakkor <edward.partenie@gmail.com>
-zhangkaihuo <zhangkaihuo@gmail.com>
 zhouwg <6889919+zhouwg@users.noreply.github.com>
-zhouwg <zhouwg2000@gmail.com>
 zrm <trustiosity.zrm@gmail.com>
-Ștefan-Gabriel Muscalu <legraphista@users.noreply.github.com>
 源文雨 <41315874+fumiama@users.noreply.github.com>
 Нияз Гарифзянов <112617865+garrnizon@users.noreply.github.com>
@@ -79,21 +79,13 @@ set(GGML_SANITIZE_ADDRESS   ${LLAMA_SANITIZE_ADDRESS})
 set(GGML_SANITIZE_UNDEFINED ${LLAMA_SANITIZE_UNDEFINED})
 set(GGML_ALL_WARNINGS       ${LLAMA_ALL_WARNINGS})
 set(GGML_FATAL_WARNINGS     ${LLAMA_FATAL_WARNINGS})
-
-# change the default for these ggml options
-if (NOT DEFINED GGML_LLAMAFILE)
-    set(GGML_LLAMAFILE ON)
-endif()
-
-if (NOT DEFINED GGML_CUDA_USE_GRAPHS)
-    set(GGML_CUDA_USE_GRAPHS ON)
-endif()
+set(GGML_LLAMAFILE          ON)

 # transition helpers
 function (llama_option_depr TYPE OLD NEW)
    if (${OLD})
        message(${TYPE} "${OLD} is deprecated and will be removed in the future.\nUse ${NEW} instead\n")
-        set(${NEW} ON PARENT_SCOPE)
+        set(${NEW} ON)
    endif()
 endfunction()

@@ -103,6 +95,7 @@ llama_option_depr(WARNING     LLAMA_KOMPUTE             GGML_KOMPUTE)
 llama_option_depr(WARNING     LLAMA_METAL               GGML_METAL)
 llama_option_depr(WARNING     LLAMA_METAL_EMBED_LIBRARY GGML_METAL_EMBED_LIBRARY)
 llama_option_depr(WARNING     LLAMA_NATIVE              GGML_NATIVE)
+llama_option_depr(WARNING     LLAMA_OPENMP              GGML_OPENMP)
 llama_option_depr(WARNING     LLAMA_RPC                 GGML_RPC)
 llama_option_depr(WARNING     LLAMA_SYCL                GGML_SYCL)
 llama_option_depr(WARNING     LLAMA_SYCL_F16            GGML_SYCL_F16)
@@ -19,7 +19,6 @@
        "cacheVariables": {
            "CMAKE_EXPORT_COMPILE_COMMANDS": "ON",
            "CMAKE_CXX_COMPILER": "icx",
-            "CMAKE_C_COMPILER": "cl",
            "GGML_SYCL": "ON",
            "CMAKE_INSTALL_RPATH": "$ORIGIN;$ORIGIN/.."
        }
@@ -45,7 +45,6 @@ BUILD_TARGETS = \
 TEST_TARGETS = \
 	tests/test-autorelease \
 	tests/test-backend-ops \
-	tests/test-chat-template \
 	tests/test-double-float \
 	tests/test-grad0 \
 	tests/test-grammar-integration \
@@ -62,11 +61,6 @@ TEST_TARGETS = \
 	tests/test-tokenizer-1-bpe \
 	tests/test-tokenizer-1-spm

-# Legacy build targets that were renamed in #7809, but should still be removed when the project is cleaned
-LEGACY_TARGETS = main quantize quantize-stats perplexity imatrix embedding vdot q8dot train-text-from-scratch convert-llama2c-to-ggml \
-	simple batched batched-bench save-load-state server gguf gguf-split eval-callback llama-bench libllava.a llava-cli baby-llama \
-	retrieval speculative infill tokenize benchmark-matmult parallel finetune export-lora lookahead lookup passkey gritlm
-
 # Deprecation aliases
 ifdef LLAMA_CUBLAS
 $(error LLAMA_CUBLAS is removed. Use GGML_CUDA instead.)
@@ -154,6 +148,12 @@ ifndef UNAME_M
 UNAME_M := $(shell uname -m)
 endif

+MK_CFLAGS     += -O3
+MK_CXXFLAGS   += -O3
+ifndef LLAMA_DEBUG
+MK_NVCCFLAGS  += -O3
+endif # LLAMA_DEBUG
+
 # In GNU make default CXX is g++ instead of c++.  Let's fix that so that users
 # of non-gcc compilers don't have to provide g++ alias or wrapper.
 DEFCC  := cc
@@ -312,10 +312,7 @@ ifdef LLAMA_DEBUG
 		MK_CPPFLAGS += -D_GLIBCXX_ASSERTIONS
 	endif
 else
-	MK_CPPFLAGS   += -DNDEBUG
-	MK_CFLAGS     += -O3
-	MK_CXXFLAGS   += -O3
-	MK_NVCCFLAGS  += -O3
+	MK_CPPFLAGS += -DNDEBUG
 endif

 ifdef LLAMA_SANITIZE_THREAD
@@ -1076,7 +1073,6 @@ clean:
 	rm -rvf src/*.o
 	rm -rvf tests/*.o
 	rm -rvf examples/*.o
-	rm -rvf common/*.o
 	rm -rvf *.a
 	rm -rvf *.dll
 	rm -rvf *.so
@@ -1091,7 +1087,6 @@ clean:
 	rm -vrf ggml/src/ggml-cuda/template-instances/*.o
 	rm -rvf $(BUILD_TARGETS)
 	rm -rvf $(TEST_TARGETS)
-	rm -rvf $(LEGACY_TARGETS)
 	find examples pocs -type f -name "*.o" -delete

 #
@@ -15,7 +15,6 @@ Inference of Meta's [LLaMA](https://arxiv.org/abs/2302.13971) model (and others)

 ### Recent API changes

- [2024 Jun 26] The source code and CMake build scripts have been restructured https://github.com/ggerganov/llama.cpp/pull/8006
 - [2024 Apr 21] `llama_token_to_piece` can now optionally render special tokens https://github.com/ggerganov/llama.cpp/pull/6807
 - [2024 Apr 4] State and session file functions reorganized under `llama_state_*` https://github.com/ggerganov/llama.cpp/pull/6341
 - [2024 Mar 26] Logits and embeddings API updated for compactness https://github.com/ggerganov/llama.cpp/pull/6122
@@ -108,7 +107,6 @@ Typically finetunes of the base models below are supported as well.
 - [X] [Falcon](https://huggingface.co/models?search=tiiuae/falcon)
 - [X] [Chinese LLaMA / Alpaca](https://github.com/ymcui/Chinese-LLaMA-Alpaca) and [Chinese LLaMA-2 / Alpaca-2](https://github.com/ymcui/Chinese-LLaMA-Alpaca-2)
 - [X] [Vigogne (French)](https://github.com/bofenghuang/vigogne)
- [X] [BERT](https://github.com/ggerganov/llama.cpp/pull/5423)
 - [X] [Koala](https://bair.berkeley.edu/blog/2023/04/03/koala/)
 - [X] [Baichuan 1 & 2](https://huggingface.co/models?search=baichuan-inc/Baichuan) + [derivations](https://huggingface.co/hiyouga/baichuan-7b-sft)
 - [X] [Aquila 1 & 2](https://huggingface.co/models?search=BAAI/Aquila)
@@ -218,11 +216,6 @@ Unless otherwise noted these projects are open-source with permissive licensing:
 **Tools:**

 - [akx/ggify](https://github.com/akx/ggify) – download PyTorch models from HuggingFace Hub and convert them to GGML
- [crashr/gppm](https://github.com/crashr/gppm) – launch llama.cpp instances utilizing NVIDIA Tesla P40 or P100 GPUs with reduced idle power consumption
-
-**Infrastructure:**
-
- [Paddler](https://github.com/distantmagic/paddler) - Stateful load balancer custom-tailored for llama.cpp

 ---

@@ -757,7 +757,7 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
        params.cache_type_v = argv[++i];
        return true;
    }
-    if (arg == "-mli" || arg == "--multiline-input") {
+    if (arg == "--multiline-input") {
        params.multiline_input = true;
        return true;
    }
@@ -1014,23 +1014,16 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
    }
    if (arg == "--in-prefix-bos") {
        params.input_prefix_bos = true;
-        params.enable_chat_template = false;
        return true;
    }
    if (arg == "--in-prefix") {
        CHECK_ARG
        params.input_prefix = argv[i];
-        params.enable_chat_template = false;
        return true;
    }
    if (arg == "--in-suffix") {
        CHECK_ARG
        params.input_suffix = argv[i];
-        params.enable_chat_template = false;
-        return true;
-    }
-    if (arg == "--spm-infill") {
-        params.spm_infill = true;
        return true;
    }
    if (arg == "--grammar") {
@@ -1409,15 +1402,13 @@ void gpt_params_print_usage(int /*argc*/, char ** argv, const gpt_params & param
                                                                        "halt generation at PROMPT, return control in interactive mode\n"
                                                                        "can be specified more than once for multiple prompts" });
    options.push_back({ "main",        "-sp,   --special",              "special tokens output enabled (default: %s)", params.special ? "true" : "false" });
-    options.push_back({ "main",        "-cnv,  --conversation",         "run in conversation mode (does not print special tokens and suffix/prefix, use default chat template) (default: %s)", params.conversation ? "true" : "false" });
+    options.push_back({ "main",        "-cnv,  --conversation",         "run in conversation mode (does not print special tokens and suffix/prefix) (default: %s)", params.conversation ? "true" : "false" });
    options.push_back({ "main infill", "-i,    --interactive",          "run in interactive mode (default: %s)", params.interactive ? "true" : "false" });
    options.push_back({ "main infill", "-if,   --interactive-first",    "run in interactive mode and wait for input right away (default: %s)", params.interactive_first ? "true" : "false" });
    options.push_back({ "main infill", "-mli,  --multiline-input",      "allows you to write or paste multiple lines without ending each in '\\'" });
    options.push_back({ "main infill", "       --in-prefix-bos",        "prefix BOS to user inputs, preceding the `--in-prefix` string" });
    options.push_back({ "main infill", "       --in-prefix STRING",     "string to prefix user inputs with (default: empty)" });
    options.push_back({ "main infill", "       --in-suffix STRING",     "string to suffix after user inputs with (default: empty)" });
-    options.push_back({ "server infill",
-                                       "       --spm-infill",           "use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: %s)", params.spm_infill ? "enabled" : "disabled" });

    options.push_back({ "sampling" });
    options.push_back({ "*",           "       --samplers SAMPLERS",    "samplers that will be used for generation in the order, separated by \';\'\n"
@@ -2627,7 +2618,6 @@ std::string llama_chat_apply_template(const struct llama_model * model,
        const std::vector<llama_chat_msg> & msgs,
        bool add_ass) {
    int alloc_size = 0;
-    bool fallback = false; // indicate if we must fallback to default chatml
    std::vector<llama_chat_message> chat;
    for (auto & msg : msgs) {
        chat.push_back({msg.role.c_str(), msg.content.c_str()});
@@ -2640,26 +2630,10 @@ std::string llama_chat_apply_template(const struct llama_model * model,
    // run the first time to get the total output length
    int32_t res = llama_chat_apply_template(model, ptr_tmpl, chat.data(), chat.size(), add_ass, buf.data(), buf.size());

-    // error: chat template is not supported
-    if (res < 0) {
-        if (ptr_tmpl != nullptr) {
-            // if the custom "tmpl" is not supported, we throw an error
-            // this is a bit redundant (for good), since we're not sure if user validated the custom template with llama_chat_verify_template()
-            throw std::runtime_error("this custom template is not supported");
-        } else {
-            // If the built-in template is not supported, we default to chatml
-            res = llama_chat_apply_template(nullptr, "chatml", chat.data(), chat.size(), add_ass, buf.data(), buf.size());
-            fallback = true;
-        }
-    }
-
    // if it turns out that our buffer is too small, we resize it
    if ((size_t) res > buf.size()) {
        buf.resize(res);
-        res = llama_chat_apply_template(
-            fallback ? nullptr : model,
-            fallback ? "chatml" : ptr_tmpl,
-            chat.data(), chat.size(), add_ass, buf.data(), buf.size());
+        res = llama_chat_apply_template(model, ptr_tmpl, chat.data(), chat.size(), add_ass, buf.data(), buf.size());
    }

    std::string formatted_chat(buf.data(), res);
@@ -2671,19 +2645,12 @@ std::string llama_chat_format_single(const struct llama_model * model,
        const std::vector<llama_chat_msg> & past_msg,
        const llama_chat_msg & new_msg,
        bool add_ass) {
-    std::ostringstream ss;
    auto fmt_past_msg = llama_chat_apply_template(model, tmpl, past_msg, false);
    std::vector<llama_chat_msg> chat_new(past_msg);
-    // if the past_msg ends with a newline, we must preserve it in the formatted version
-    if (add_ass && !fmt_past_msg.empty() && fmt_past_msg.back() == '\n') {
-        ss << "\n";
-    };
-    // format chat with new_msg
    chat_new.push_back(new_msg);
    auto fmt_new_msg = llama_chat_apply_template(model, tmpl, chat_new, add_ass);
-    // get the diff part
-    ss << fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size());
-    return ss.str();
+    auto formatted = fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size());
+    return formatted;
 }

 std::string llama_chat_format_example(const struct llama_model * model,
@@ -2837,87 +2804,125 @@ float llama_embd_similarity_cos(const float * embd1, const float * embd2, int n)
 //

 static llama_control_vector_data llama_control_vector_load_one(const llama_control_vector_load_info & load_info) {
+    int32_t n_tensors;
+
+    size_t n_bytes = 0;
+
+    uint32_t max_direction_layer = 0;
+
    llama_control_vector_data result = { -1, {} };

-    ggml_context * ctx = nullptr;
-    struct gguf_init_params meta_gguf_params = {
-        /* .no_alloc = */ false,
-        /* .ctx      = */ &ctx,
-    };
-    struct gguf_context * ctx_gguf = gguf_init_from_file(load_info.fname.c_str(), meta_gguf_params);
-    if (!ctx_gguf) {
-        fprintf(stderr, "%s: failed to load control vector file from %s\n", __func__, load_info.fname.c_str());
+    // calculate size of ctx needed for tensors, ensure tensors are f32, and find max layer
+    {
+        struct ggml_init_params meta_params = {
+            /* .mem_size   = */ ggml_tensor_overhead() * 128 + ggml_graph_overhead(),
+            /* .mem_buffer = */ nullptr,
+            /* .no_alloc   = */ true,
+        };
+        ggml_context * meta_ctx = ggml_init(meta_params);
+        struct gguf_init_params meta_gguf_params = {
+            /* .no_alloc = */ true,
+            /* .ctx      = */ &meta_ctx,
+        };
+        struct gguf_context * meta_ctx_gguf = gguf_init_from_file(load_info.fname.c_str(), meta_gguf_params);
+        if (!meta_ctx_gguf) {
+            fprintf(stderr, "%s: failed to load control vector from %s\n", __func__, load_info.fname.c_str());
+            ggml_free(meta_ctx);
+            return result;
+        }
+
+        n_tensors = gguf_get_n_tensors(meta_ctx_gguf);
+        for (int i = 0; i < n_tensors; i++) {
+            std::string name = gguf_get_tensor_name(meta_ctx_gguf, i);
+
+            // split on '.'
+            size_t dotpos = name.find('.');
+            if (dotpos != std::string::npos && name.substr(0, dotpos) == "direction") {
+                try {
+                    uint32_t layer = std::stoi(name.substr(dotpos + 1));
+                    if (layer == 0) {
+                        fprintf(stderr, "%s: direction tensor invalid in %s\n", __func__, load_info.fname.c_str());
+                        ggml_free(meta_ctx);
+                        gguf_free(meta_ctx_gguf);
+                        return result;
+                    }
+                    if (layer > max_direction_layer) {
+                        max_direction_layer = layer;
+                    }
+                } catch (...) {
+                    fprintf(stderr, "%s: direction tensor invalid in %s\n", __func__, load_info.fname.c_str());
+                    ggml_free(meta_ctx);
+                    gguf_free(meta_ctx_gguf);
+                    return result;
+                }
+            }
+
+            struct ggml_tensor * tensor_meta = ggml_get_tensor(meta_ctx, name.c_str());
+            if (tensor_meta->type != GGML_TYPE_F32 || ggml_n_dims(tensor_meta) != 1) {
+                fprintf(stderr, "%s: direction tensor invalid in %s\n", __func__, load_info.fname.c_str());
+                ggml_free(meta_ctx);
+                gguf_free(meta_ctx_gguf);
+                return result;
+            }
+            if (result.n_embd == -1) {
+                result.n_embd = ggml_nelements(tensor_meta);
+            } else if (ggml_nelements(tensor_meta) != result.n_embd) {
+                fprintf(stderr, "%s: direction tensor sizes mismatched in %s\n", __func__, load_info.fname.c_str());
+                ggml_free(meta_ctx);
+                gguf_free(meta_ctx_gguf);
+                return result;
+            }
+            n_bytes += ggml_nbytes(tensor_meta);
+        }
+        ggml_free(meta_ctx);
+        gguf_free(meta_ctx_gguf);
+    }
+
+    if (n_tensors == 0) {
+        fprintf(stderr, "%s: no direction tensors found in %s\n", __func__, load_info.fname.c_str());
        return result;
    }

-    int32_t n_tensors = gguf_get_n_tensors(ctx_gguf);
-    if (n_tensors == 0) {
-        fprintf(stderr, "%s: no direction tensors found in %s\n", __func__, load_info.fname.c_str());
+    // load and scale tensors into final control vector context
+    struct ggml_init_params ggml_params = {
+        /* .mem_size   = */ ggml_tensor_overhead() * n_tensors + n_bytes,
+        /* .mem_buffer = */ nullptr,
+        /* .no_alloc   = */ false,
+    };
+    struct ggml_context * ctx = ggml_init(ggml_params);
+
+    struct gguf_init_params params = {
+        /*.no_alloc = */ false,
+        /*.ctx      = */ &ctx,
+    };
+    struct gguf_context * ctx_gguf = gguf_init_from_file(load_info.fname.c_str(), params);
+    if (!ctx_gguf) {
+        fprintf(stderr, "%s: failed to load control vector from %s\n", __func__, load_info.fname.c_str());
+        ggml_free(ctx);
+        return result;
    }

-    for (int i = 0; i < n_tensors; i++) {
-        std::string name = gguf_get_tensor_name(ctx_gguf, i);
+    // do not store data for layer 0 (it's not used)
+    result.data.resize(result.n_embd * max_direction_layer);

-        int layer_idx = -1;
+    for (uint32_t il = 1; il <= max_direction_layer; il++) {
+        const std::string name = "direction." + std::to_string(il);
+        const ggml_tensor * tensor = ggml_get_tensor(ctx, name.c_str());

-        // split on '.'
-        size_t dotpos = name.find('.');
-        if (dotpos != std::string::npos && name.substr(0, dotpos) == "direction") {
-            try {
-                layer_idx = std::stoi(name.substr(dotpos + 1));
-            } catch (...) {
-                layer_idx = -1;
+        float * dst = result.data.data() + result.n_embd * (il - 1);
+
+        if (tensor) {
+            const float * src = (const float *) tensor->data;
+            for (int j = 0; j < result.n_embd; j++) {
+                dst[j] = src[j] * load_info.strength;
+            }
+        } else {
+            for (int j = 0; j < result.n_embd; j++) {
+                dst[j] = 0.0f;
            }
        }
-        if (layer_idx < 0) {
-            fprintf(stderr, "%s: invalid/unparsable direction tensor layer index in %s\n", __func__, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
-        } else if (layer_idx == 0) {
-            fprintf(stderr, "%s: invalid (zero) direction tensor layer index in %s\n", __func__, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
-        }
-
-        struct ggml_tensor * tensor = ggml_get_tensor(ctx, name.c_str());
-        if (tensor->type != GGML_TYPE_F32) {
-            fprintf(stderr, "%s: invalid (non-F32) direction tensor type in %s\n", __func__, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
-        }
-        if (ggml_n_dims(tensor) != 1) {
-            fprintf(stderr, "%s: invalid (non-1D) direction tensor shape in %s\n", __func__, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
-        }
-
-        if (result.n_embd == -1) {
-            result.n_embd = ggml_nelements(tensor);
-        } else if (ggml_nelements(tensor) != result.n_embd) {
-            fprintf(stderr, "%s: direction tensor in %s does not match previous dimensions\n", __func__, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
-        }
-
-        // extend if necessary - do not store data for layer 0 (it's not used)
-        result.data.resize(std::max(result.data.size(), static_cast<size_t>(result.n_embd * layer_idx)), 0.0f);
-
-        const float * src = (const float *) tensor->data;
-        float * dst = result.data.data() + result.n_embd * (layer_idx - 1);  // layer 1 at [0]
-        for (int j = 0; j < result.n_embd; j++) {
-            dst[j] += src[j] * load_info.strength;  // allows multiple directions for same layer in same file
-        }
-
    }

-    if (result.n_embd == -1) {
-        fprintf(stderr, "%s: skipping %s due to invalid direction tensors\n", __func__, load_info.fname.c_str());
-        result.data.clear();
-    }
-
-    gguf_free(ctx_gguf);
-    ggml_free(ctx);
-
    return result;
 }

@@ -2928,19 +2933,16 @@ llama_control_vector_data llama_control_vector_load(const std::vector<llama_cont
        auto cur = llama_control_vector_load_one(info);

        if (cur.n_embd == -1) {
-            result.n_embd = -1;
-            break;
+            return result;
        }
-        if (result.n_embd != -1 && result.n_embd != cur.n_embd) {
-            fprintf(stderr, "%s: control vectors in %s does not match previous dimensions\n", __func__, info.fname.c_str());
-            result.n_embd = -1;
-            break;
+        if (result.n_embd != -1 && (result.n_embd != cur.n_embd || result.data.size() != cur.data.size())) {
+            fprintf(stderr, "%s: control vector in %s does not match previous vector dimensions\n", __func__, info.fname.c_str());
+            return result;
        }

        if (result.n_embd == -1) {
            result = std::move(cur);
        } else {
-            result.data.resize(std::max(result.data.size(), cur.data.size()), 0.0f);  // extend if necessary
            for (size_t i = 0; i < cur.data.size(); i++) {
                result.data[i] += cur.data[i];
            }
@@ -2948,8 +2950,7 @@ llama_control_vector_data llama_control_vector_load(const std::vector<llama_cont
    }

    if (result.n_embd == -1) {
-        fprintf(stderr, "%s: no valid control vector files passed\n", __func__);
-        result.data.clear();
+        fprintf(stderr, "%s: no vectors passed\n", __func__);
    }

    return result;
@@ -200,7 +200,6 @@ struct gpt_params {
    std::string public_path   = "";
    std::string chat_template = "";
    std::string system_prompt = "";
-    bool enable_chat_template = true;

    std::vector<std::string> api_keys;

@@ -251,8 +250,6 @@ struct gpt_params {
    std::string cvector_outfile       = "control_vector.gguf";
    std::string cvector_positive_file = "examples/cvector-generator/positive.txt";
    std::string cvector_negative_file = "examples/cvector-generator/negative.txt";
-
-    bool spm_infill = false; // suffix/prefix/middle pattern for infill
 };

 void gpt_params_handle_model_default(gpt_params & params);
@@ -383,8 +380,6 @@ struct llama_chat_msg {
 bool llama_chat_verify_template(const std::string & tmpl);

 // CPP wrapper for llama_chat_apply_template
-// If the built-in template is not supported, we default to chatml
-// If the custom "tmpl" is not supported, we throw an error
 std::string llama_chat_apply_template(const struct llama_model * model,
        const std::string & tmpl,
        const std::vector<llama_chat_msg> & chat,
@@ -459,3 +454,4 @@ void yaml_dump_string_multiline(FILE * stream, const char * prop_name, const cha
 void yaml_dump_non_result_info(
    FILE * stream, const gpt_params & params, const llama_context * lctx,
    const std::string & timestamp, const std::vector<int> & prompt_tokens, const char * model_desc);
+
@@ -316,7 +316,7 @@ std::unordered_map<char, std::string> GRAMMAR_LITERAL_ESCAPES = {
 };

 std::unordered_set<char> NON_LITERAL_SET = {'|', '.', '(', ')', '[', ']', '{', '}', '*', '+', '?'};
-std::unordered_set<char> ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = {'^', '$', '.', '[', ']', '(', ')', '|', '{', '}', '*', '+', '?'};
+std::unordered_set<char> ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = {'[', ']', '(', ')', '|', '{', '}', '*', '+', '?'};

 template <typename Iterator>
 std::string join(Iterator begin, Iterator end, const std::string & separator) {
@@ -720,7 +720,7 @@ private:
            }
            prop_names.push_back(prop_name);
        }
-        if ((additional_properties.is_boolean() && additional_properties.get<bool>()) || additional_properties.is_object()) {
+        if (!(additional_properties.is_boolean() && !additional_properties.get<bool>())) {
            std::string sub_name = name + (name.empty() ? "" : "-") + "additional";
            std::string value_rule =
                additional_properties.is_object() ? visit(additional_properties, sub_name + "-value")
@@ -85,8 +85,6 @@ models = [
    {"name": "smaug-bpe",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/abacusai/Smaug-Llama-3-70B-Instruct", },
    {"name": "poro-chat",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Poro-34B-chat", },
    {"name": "jina-v2-code",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-code", },
-    {"name": "viking",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Viking-7B", }, # Also used for Viking 13B and 33B
-    {"name": "jais",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/core42/jais-13b", },
 ]


@@ -487,12 +487,6 @@ class Model:
        if chkhsh == "7967bfa498ade6b757b064f31e964dddbb80f8f9a4d68d4ba7998fcf281c531a":
            # ref: https://huggingface.co/jinaai/jina-embeddings-v2-base-code
            res = "jina-v2-code"
-        if chkhsh == "7fc505bd3104ca1083b150b17d088b59534ede9bde81f0dd2090967d7fe52cee":
-            # ref: https://huggingface.co/LumiOpen/Viking-7B
-            res = "viking"
-        if chkhsh == "b53802fb28e26d645c3a310b34bfe07da813026ec7c7716883404d5e0f8b1901":
-            # ref: https://huggingface.co/core42/jais-13b
-            res = "jais"

        if res is None:
            logger.warning("\n")
@@ -579,19 +573,7 @@ class Model:
        special_vocab._set_special_token("unk", tokenizer.special_tokens["<|endoftext|>"])
        special_vocab.add_to_gguf(self.gguf_writer)

-    def _set_vocab_sentencepiece(self, add_to_gguf=True):
-        tokens, scores, toktypes = self._create_vocab_sentencepiece()
-
-        self.gguf_writer.add_tokenizer_model("llama")
-        self.gguf_writer.add_tokenizer_pre("default")
-        self.gguf_writer.add_token_list(tokens)
-        self.gguf_writer.add_token_scores(scores)
-        self.gguf_writer.add_token_types(toktypes)
-
-        special_vocab = gguf.SpecialVocab(self.dir_model, n_vocab=len(tokens))
-        special_vocab.add_to_gguf(self.gguf_writer)
-
-    def _create_vocab_sentencepiece(self):
+    def _set_vocab_sentencepiece(self):
        from sentencepiece import SentencePieceProcessor

        tokenizer_path = self.dir_model / 'tokenizer.model'
@@ -653,7 +635,14 @@ class Model:
                scores.append(-1000.0)
                toktypes.append(SentencePieceTokenTypes.UNUSED)

-        return tokens, scores, toktypes
+        self.gguf_writer.add_tokenizer_model("llama")
+        self.gguf_writer.add_tokenizer_pre("default")
+        self.gguf_writer.add_token_list(tokens)
+        self.gguf_writer.add_token_scores(scores)
+        self.gguf_writer.add_token_types(toktypes)
+
+        special_vocab = gguf.SpecialVocab(self.dir_model, n_vocab=len(tokens))
+        special_vocab.add_to_gguf(self.gguf_writer)

    def _set_vocab_llama_hf(self):
        vocab = gguf.LlamaHfVocab(self.dir_model)
@@ -2348,70 +2337,6 @@ class GemmaModel(Model):
        return [(self.map_tensor_name(name), data_torch)]


-@Model.register("Gemma2ForCausalLM")
-class Gemma2Model(Model):
-    model_arch = gguf.MODEL_ARCH.GEMMA2
-
-    def set_vocab(self):
-        tokens, scores, toktypes = self._create_vocab_sentencepiece()
-        # hack: This is required so that we can properly use start/end-of-turn for chat template
-        for i in range(108):
-            # including <unusedX>, <start_of_turn>, <end_of_turn>
-            toktypes[i] = SentencePieceTokenTypes.CONTROL
-        self.gguf_writer.add_tokenizer_model("llama")
-        self.gguf_writer.add_tokenizer_pre("default")
-        self.gguf_writer.add_token_list(tokens)
-        self.gguf_writer.add_token_scores(scores)
-        self.gguf_writer.add_token_types(toktypes)
-
-        special_vocab = gguf.SpecialVocab(self.dir_model, n_vocab=len(tokens))
-        special_vocab.add_to_gguf(self.gguf_writer)
-        self.gguf_writer.add_add_space_prefix(False)
-
-    def set_gguf_parameters(self):
-        hparams = self.hparams
-        block_count = hparams["num_hidden_layers"]
-
-        self.gguf_writer.add_name(self.dir_model.name if self.model_name is None else self.model_name)
-        self.gguf_writer.add_context_length(hparams["max_position_embeddings"])
-        self.gguf_writer.add_embedding_length(hparams["hidden_size"])
-        self.gguf_writer.add_block_count(block_count)
-        self.gguf_writer.add_feed_forward_length(hparams["intermediate_size"])
-        self.gguf_writer.add_head_count(hparams["num_attention_heads"])
-        self.gguf_writer.add_head_count_kv(self.hparams["num_key_value_heads"] if "num_key_value_heads" in hparams else hparams["num_attention_heads"])
-        self.gguf_writer.add_layer_norm_rms_eps(self.hparams["rms_norm_eps"])
-        self.gguf_writer.add_key_length(hparams["head_dim"])
-        self.gguf_writer.add_value_length(hparams["head_dim"])
-        self.gguf_writer.add_file_type(self.ftype)
-        self.gguf_writer.add_attn_logit_softcapping(
-            self.hparams["attn_logit_softcapping"]
-        )
-        self.gguf_writer.add_final_logit_softcapping(
-            self.hparams["final_logit_softcapping"]
-        )
-        self.gguf_writer.add_sliding_window(self.hparams["sliding_window"])
-
-        # sanity check
-        attn_scalar = self.hparams["query_pre_attn_scalar"]
-        if attn_scalar != hparams["hidden_size"] / hparams["num_attention_heads"]:
-            raise ValueError("query_pre_attn_scalar must be equal to n_embd / n_head")
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        del bid  # unusem
-
-        # lm_head is not used in llama.cpp, while autoawq will include this tensor in model
-        # To prevent errors, skip loading lm_head.weight.
-        if name == "lm_head.weight":
-            logger.debug(f"Skipping get tensor {name!r} in safetensors so that convert can end normally.")
-            return []
-
-        # ref: https://github.com/huggingface/transformers/blob/fc37f38915372c15992b540dfcbbe00a916d4fc6/src/transformers/models/gemma/modeling_gemma.py#L89
-        if name.endswith("norm.weight"):
-            data_torch = data_torch + 1
-
-        return [(self.map_tensor_name(name), data_torch)]
-
-
@Model.register("Starcoder2ForCausalLM")
 class StarCoder2Model(Model):
    model_arch = gguf.MODEL_ARCH.STARCODER2
@@ -2968,96 +2893,6 @@ class T5Model(Model):
        return [(self.map_tensor_name(name), data_torch)]


-@Model.register("JAISLMHeadModel")
-class JaisModel(Model):
-    model_arch = gguf.MODEL_ARCH.JAIS
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-
-        # SwigLU activation
-        assert self.hparams["activation_function"] == "swiglu"
-        # ALiBi position embedding
-        assert self.hparams["position_embedding_type"] == "alibi"
-
-        # Embeddings scale
-        self.embeddings_scale = 1.0
-        # note: For some JAIS flavors, output is tied to (same as) wte in original model
-        self.output_is_wte = False
-        if 'mup_embeddings_scale' in self.hparams:
-            self.output_is_wte = True   # Hack (?)
-            self.embeddings_scale = self.hparams['mup_embeddings_scale']
-        elif 'embeddings_scale' in self.hparams:
-            self.embeddings_scale = self.hparams['embeddings_scale']
-        else:
-            assert False
-
-        self.width_scale = 1.0
-        if 'mup_output_alpha' in self.hparams:
-            assert 'mup_width_scale' in self.hparams
-            self.width_scale = self.hparams['mup_output_alpha'] * self.hparams['mup_width_scale']
-        elif 'width_scale' in self.hparams:
-            self.width_scale = self.hparams['width_scale']
-        else:
-            assert False
-
-        self.max_alibi_bias = 8.0
-
-    def set_vocab(self):
-        self._set_vocab_gpt2()
-
-    def set_gguf_parameters(self):
-        self.gguf_writer.add_name(self.dir_model.name)
-        self.gguf_writer.add_block_count(self.hparams["n_layer"])
-        self.gguf_writer.add_context_length(self.hparams["n_positions"])
-        self.gguf_writer.add_embedding_length(self.hparams["n_embd"])
-        self.gguf_writer.add_feed_forward_length(self.hparams["n_inner"])
-        self.gguf_writer.add_head_count(self.hparams["n_head"])
-        self.gguf_writer.add_layer_norm_eps(self.hparams["layer_norm_epsilon"])
-        self.gguf_writer.add_file_type(self.ftype)
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        del bid  # unused
-
-        tensors: list[tuple[str, Tensor]] = []
-
-        # we don't need these
-        if name.endswith((".attn.bias")):
-            return tensors
-
-        if name.endswith(("relative_pe.slopes")):
-            # Calculate max ALiBi bias (this is the inverse of the ALiBi calculation)
-            # Some other models has max_alibi_bias spelled out explicitly in the hyperparams,
-            # but Jais's PyTorch model simply precalculates the slope values and places them
-            # in relative_pes.slopes
-            n_head_closest_log2 = 2 ** math.floor(math.log2(self.hparams["n_head"]))
-            first_val = float(data_torch._data[0])
-            self.max_alibi_bias = -round(math.log2(first_val) * n_head_closest_log2)
-
-            return tensors
-
-        if name.endswith((".c_attn.weight", ".c_proj.weight", ".c_fc.weight", ".c_fc2.weight")):
-            data_torch = data_torch.transpose(1, 0)
-
-        new_name = self.map_tensor_name(name)
-
-        if new_name == self.format_tensor_name(gguf.MODEL_TENSOR.TOKEN_EMBD):
-            tensors.append((new_name, data_torch * self.embeddings_scale))
-            if self.output_is_wte:
-                tensors.append((self.format_tensor_name(gguf.MODEL_TENSOR.OUTPUT), data_torch * self.width_scale))
-        elif new_name == self.format_tensor_name(gguf.MODEL_TENSOR.OUTPUT):
-            assert not self.output_is_wte
-            tensors.append((new_name, data_torch * self.width_scale))
-        else:
-            tensors.append((new_name, data_torch))
-
-        return tensors
-
-    def write_tensors(self):
-        super().write_tensors()
-        self.gguf_writer.add_max_alibi_bias(self.max_alibi_bias)
-
-
 ###### CONVERSION LOGIC ######


@@ -3213,8 +3048,7 @@ def main() -> None:
        "auto": gguf.LlamaFileType.GUESSED,
    }

-    is_split = args.split_max_tensors > 0 or args.split_max_size != "0"
-    if args.use_temp_file and is_split:
+    if args.use_temp_file and (args.split_max_tensors > 0 or args.split_max_size != "0"):
        logger.error("Error: Cannot use temp file when splitting")
        sys.exit(1)

@@ -3251,12 +3085,11 @@ def main() -> None:
        if args.vocab_only:
            logger.info("Exporting model vocab...")
            model_instance.write_vocab()
-            logger.info(f"Model vocab successfully exported to {model_instance.fname_out}")
+            logger.info("Model vocab successfully exported.")
        else:
            logger.info("Exporting model...")
            model_instance.write()
-            out_path = f"{model_instance.fname_out.parent}{os.sep}" if is_split else model_instance.fname_out
-            logger.info(f"Model successfully exported to {out_path}")
+            logger.info("Model successfully exported.")


 if __name__ == '__main__':
@@ -58,3 +58,4 @@ The above command will output space-separated float values.
 ```powershell
 embedding.exe -p 'Castle<#sep#>Stronghold<#sep#>Dog<#sep#>Cat' --embd-separator '<#sep#>' --embd-normalize 2  --embd-output-format '' -m './path/to/model.gguf' --n-gpu-layers 99 --log-disable 2>/dev/null
 ```
+
@@ -15,7 +15,6 @@ In this section, we cover the most commonly used options for running the `infill
 -   `-i, --interactive`: Run the program in interactive mode, allowing you to provide input directly and receive real-time responses.
 -   `-n N, --n-predict N`: Set the number of tokens to predict when generating text. Adjusting this value can influence the length of the generated text.
 -   `-c N, --ctx-size N`: Set the size of the prompt context. The default is 512, but LLaMA models were built with a context of 2048, which will provide better results for longer input/inference.
-   `--spm-infill`: Use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this.

 ## Input Prompts

@@ -210,7 +210,6 @@ int main(int argc, char ** argv) {
        suff_rm_leading_spc = false;
    }
    std::vector<llama_token> embd_inp;
-    std::vector<llama_token> embd_end;
    std::vector<llama_token> inp_pfx = ::llama_tokenize(ctx, params.input_prefix, false);
    std::vector<llama_token> inp_sfx = ::llama_tokenize(ctx, params.input_suffix, false);
    const int space_token = 29871;
@@ -218,13 +217,12 @@ int main(int argc, char ** argv) {
        inp_sfx.erase(inp_sfx.begin());
    }
    inp_pfx.insert(inp_pfx.begin(), llama_token_prefix(model));
-    inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model));
-    embd_inp = params.spm_infill ? inp_sfx : inp_pfx;
-    embd_end = params.spm_infill ? inp_pfx : inp_sfx;
    if (add_bos) {
-        embd_inp.insert(embd_inp.begin(), llama_token_bos(model));
+        inp_pfx.insert(inp_pfx.begin(), llama_token_bos(model));
    }
-    embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end());
+    inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model));
+    embd_inp = inp_pfx;
+    embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end());

    const llama_token middle_token = llama_token_middle(model);
    if (middle_token >= 0) {
@@ -528,14 +526,14 @@ int main(int argc, char ** argv) {
                    inp_sfx.erase(inp_sfx.begin());
                }
                inp_pfx.insert(inp_pfx.begin(), llama_token_prefix(model));
-                inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model));
-                embd_inp = params.spm_infill ? inp_sfx : inp_pfx;
-                embd_end = params.spm_infill ? inp_pfx : inp_sfx;
                if (add_bos) {
-                    embd_inp.insert(embd_inp.begin(), llama_token_bos(model));
+                    inp_pfx.insert(inp_pfx.begin(), llama_token_bos(model));
                }
-                embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end());
+                inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model));
+                embd_inp = inp_pfx;
+                embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end());

+                const llama_token middle_token = llama_token_middle(model);
                if (middle_token >= 0) {
                    embd_inp.push_back(middle_token);
                }
@@ -659,3 +657,4 @@ int main(int argc, char ** argv) {

    return 0;
 }
+
@@ -231,7 +231,7 @@ GRAMMAR_RANGE_LITERAL_ESCAPE_RE = re.compile(r'[\r\n"\]\-\\]')
 GRAMMAR_LITERAL_ESCAPES = {'\r': '\\r', '\n': '\\n', '"': '\\"', '-': '\\-', ']': '\\]'}

 NON_LITERAL_SET = set('|.()[]{}*+?')
-ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = set('^$.[]()|{}*+?')
+ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = set('[]()|{}*+?')


 class SchemaConverter:
@@ -602,7 +602,7 @@ class SchemaConverter:
                else:
                    add_component(t, is_required=True)

-            return self._add_rule(rule_name, self._build_object_rule(properties, required, hybrid_name, additional_properties=None))
+            return self._add_rule(rule_name, self._build_object_rule(properties, required, hybrid_name, additional_properties=[]))

        elif schema_type in (None, 'array') and ('items' in schema or 'prefixItems' in schema):
            items = schema.get('items') or schema['prefixItems']
@@ -691,7 +691,7 @@ class SchemaConverter:
        required_props = [k for k in sorted_props if k in required]
        optional_props = [k for k in sorted_props if k not in required]

-        if additional_properties is not None and additional_properties != False:
+        if additional_properties != False:
            sub_name = f'{name}{"-" if name else ""}additional'
            value_rule = self.visit(additional_properties, f'{sub_name}-value') if isinstance(additional_properties, dict) else \
                self._add_primitive('value', PRIMITIVE_RULES['value'])
@@ -0,0 +1,55 @@
+
+# For more information about using CMake with Android Studio, read the
+# documentation: https://d.android.com/studio/projects/add-native-code.html.
+# For more examples on how to use CMake, see https://github.com/android/ndk-samples.
+
+# Sets the minimum CMake version required for this project.
+cmake_minimum_required(VERSION 3.22.1)
+
+# Declares the project name. The project name can be accessed via ${ PROJECT_NAME},
+# Since this is the top level CMakeLists.txt, the project name is also accessible
+# with ${CMAKE_PROJECT_NAME} (both CMake variables are in-sync within the top level
+# build script scope).
+project("llama-android")
+
+## Fetch latest llama.cpp from GitHub
+#include(FetchContent)
+#FetchContent_Declare(
+#        llama
+#        GIT_REPOSITORY https://github.com/ggerganov/llama.cpp
+#        GIT_TAG        master
+#)
+#
+## Also provides "common"
+#FetchContent_MakeAvailable(llama)
+
+# llama.cpp CI uses the code from the current branch
+# ref: https://github.com/ggerganov/llama.cpp/pull/7341#issuecomment-2117617700
+add_subdirectory(../../../../../../ build-llama)
+
+# Creates and names a library, sets it as either STATIC
+# or SHARED, and provides the relative paths to its source code.
+# You can define multiple libraries, and CMake builds them for you.
+# Gradle automatically packages shared libraries with your APK.
+#
+# In this top level CMakeLists.txt, ${CMAKE_PROJECT_NAME} is used to define
+# the target library name; in the sub-module's CMakeLists.txt, ${PROJECT_NAME}
+# is preferred for the same purpose.
+#
+# In order to load a library into your app from Java/Kotlin, you must call
+# System.loadLibrary() and pass the name of the library defined here;
+# for GameActivity/NativeActivity derived applications, the same library name must be
+# used in the AndroidManifest.xml file.
+add_library(${CMAKE_PROJECT_NAME} SHARED
+    # List C/C++ source files with relative paths to this CMakeLists.txt.
+        llama-android.cpp)
+
+# Specifies libraries CMake should link to your target library. You
+# can link libraries from various origins, such as libraries defined in this
+# build script, prebuilt third-party libraries, or Android system libraries.
+target_link_libraries(${CMAKE_PROJECT_NAME}
+    # List libraries link to the target library
+    llama
+    common
+    android
+    log)
@@ -11,15 +11,15 @@ cmake_minimum_required(VERSION 3.22.1)
 # build script scope).
 project("llama-android")

-#include(FetchContent)
-#FetchContent_Declare(
-#        llama
-#        GIT_REPOSITORY https://github.com/ggerganov/llama.cpp
-#        GIT_TAG        master
-#)
+include(FetchContent)
+FetchContent_Declare(
+        llama
+        GIT_REPOSITORY https://github.com/ggerganov/llama.cpp
+        GIT_TAG        master
+)

 # Also provides "common"
-#FetchContent_MakeAvailable(llama)
+FetchContent_MakeAvailable(llama)

 # Creates and names a library, sets it as either STATIC
 # or SHARED, and provides the relative paths to its source code.
@@ -30,10 +30,6 @@ project("llama-android")
 # the target library name; in the sub-module's CMakeLists.txt, ${PROJECT_NAME}
 # is preferred for the same purpose.
 #
-
-#load local llama.cpp
-add_subdirectory(../../../../../../ build-llama)
-
 # In order to load a library into your app from Java/Kotlin, you must call
 # System.loadLibrary() and pass the name of the library defined here;
 # for GameActivity/NativeActivity derived applications, the same library name must be
@@ -5,7 +5,7 @@
 #include <string>
 #include <unistd.h>
 #include "llama.h"
-#include "common.h"
+#include "common/common.h"

 // Write C++ code here.
 //
@@ -1121,20 +1121,20 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
            }
            if (n < 32)
                hparams.image_grid_pinpoints[n] = 0;
-        } catch (std::runtime_error & /*e*/) {
+        } catch (std::runtime_error & e) {
            hparams.image_grid_pinpoints[0]=0;
        }

        try {
            int idx = get_key_idx(ctx, KEY_MM_PATCH_MERGE_TYPE);
            strcpy(hparams.mm_patch_merge_type, gguf_get_val_str(ctx, idx));
-        } catch (std::runtime_error & /*e*/) {
+        } catch (std::runtime_error & e) {
            strcpy(hparams.mm_patch_merge_type, "flat");
        }

        try {
            hparams.image_crop_resolution = get_u32(ctx, KEY_IMAGE_CROP_RESOLUTION); // llava-1.6
-        } catch(const std::exception& /*e*/) {
+        } catch(const std::exception& e) {
            hparams.image_crop_resolution = hparams.image_size;
        }

@@ -1173,7 +1173,7 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
        try {
            vision_model.class_embedding  = get_tensor(new_clip->ctx_data, TN_CLASS_EMBD);
            new_clip->has_class_embedding = true;
-        } catch (const std::exception& /*e*/) {
+        } catch (const std::exception& e) {
            new_clip->has_class_embedding = false;
        }

@@ -1181,7 +1181,7 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
            vision_model.pre_ln_w  = get_tensor(new_clip->ctx_data, format(TN_LN_PRE, "v", "weight"));
            vision_model.pre_ln_b  = get_tensor(new_clip->ctx_data, format(TN_LN_PRE, "v", "bias"));
            new_clip->has_pre_norm = true;
-        } catch (std::exception & /*e*/) {
+        } catch (std::exception & e) {
            new_clip->has_pre_norm = false;
        }

@@ -1189,21 +1189,21 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
            vision_model.post_ln_w  = get_tensor(new_clip->ctx_data, format(TN_LN_POST, "v", "weight"));
            vision_model.post_ln_b  = get_tensor(new_clip->ctx_data, format(TN_LN_POST, "v", "bias"));
            new_clip->has_post_norm = true;
-        } catch (std::exception & /*e*/) {
+        } catch (std::exception & e) {
            new_clip->has_post_norm = false;
        }

        try {
            vision_model.patch_bias = get_tensor(new_clip->ctx_data, TN_PATCH_BIAS);
            new_clip->has_patch_bias = true;
-        } catch (std::exception & /*e*/) {
+        } catch (std::exception & e) {
            new_clip->has_patch_bias = false;
        }

        try {
            vision_model.patch_embeddings    = get_tensor(new_clip->ctx_data, TN_PATCH_EMBD);
            vision_model.position_embeddings = get_tensor(new_clip->ctx_data, format(TN_POS_EMBD, "v"));
-        } catch(const std::exception& /*e*/) {
+        } catch(const std::exception& e) {
            LOG_TEE("%s: failed to load vision model tensors\n", __func__);
        }

@@ -1215,26 +1215,26 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
                // Yi-type llava
                vision_model.mm_1_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 1, "weight"));
                vision_model.mm_1_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 1, "bias"));
-            } catch (std::runtime_error & /*e*/) { }
+            } catch (std::runtime_error & e) {  }
            try {
                // missing in Yi-type llava
                vision_model.mm_2_w              = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 2, "weight"));
                vision_model.mm_2_b              = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 2, "bias"));
-            } catch (std::runtime_error & /*e*/) { }
+            } catch (std::runtime_error & e) {  }
            try {
                // Yi-type llava
                vision_model.mm_3_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 3, "weight"));
                vision_model.mm_3_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 3, "bias"));
-            } catch (std::runtime_error & /*e*/) { }
+            } catch (std::runtime_error & e) {  }
            try {
                // Yi-type llava
                vision_model.mm_4_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 4, "weight"));
                vision_model.mm_4_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 4, "bias"));
-            } catch (std::runtime_error & /*e*/) { }
+            } catch (std::runtime_error & e) {  }
            try {
                vision_model.image_newline = get_tensor(new_clip->ctx_data, TN_IMAGE_NEWLINE);
                // LOG_TEE("%s: image_newline tensor (llava-1.6) found\n", __func__);
-            } catch (std::runtime_error & /*e*/) { }
+            } catch (std::runtime_error & e) {  }
        } else if (new_clip->proj_type == PROJECTOR_TYPE_LDP) {
            // MobileVLM projection
            vision_model.mm_model_mlp_1_w               = get_tensor(new_clip->ctx_data, format(TN_MVLM_PROJ_MLP, 1, "weight"));
@@ -10,3 +10,4 @@ More info:

 https://github.com/ggerganov/llama.cpp/pull/4484
 https://github.com/ggerganov/llama.cpp/issues/4226
+
@@ -48,3 +48,4 @@
 build*/
 out/
 tmp/
+
@@ -30,3 +30,4 @@ target_include_directories(${TARGET} PRIVATE ${_common_path})
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
 target_compile_features(${TARGET} PRIVATE cxx_std_11)
+
@@ -261,7 +261,7 @@ int main(int argc, char ** argv) {
    std::vector<llama_token> embd_inp;

    {
-        auto prompt = (params.conversation && params.enable_chat_template)
+        auto prompt = params.conversation
            ? chat_add_and_format(model, chat_msgs, "system", params.prompt) // format the system prompt in conversation mode
            : params.prompt;
        if (params.interactive_first || !params.prompt.empty() || session_tokens.empty()) {
@@ -810,9 +810,7 @@ int main(int argc, char ** argv) {
                        is_antiprompt = true;
                    }

-                    if (params.enable_chat_template) {
-                        chat_add_and_format(model, chat_msgs, "assistant", assistant_ss.str());
-                    }
+                    chat_add_and_format(model, chat_msgs, "system", assistant_ss.str());
                    is_interacting = true;
                    printf("\n");
                }
@@ -874,13 +872,12 @@ int main(int argc, char ** argv) {
                        string_process_escapes(buffer);
                    }

-                    bool format_chat = params.conversation && params.enable_chat_template;
-                    std::string user_inp = format_chat
+                    std::string user_inp = params.conversation
                        ? chat_add_and_format(model, chat_msgs, "user", std::move(buffer))
                        : std::move(buffer);
                    // TODO: one inconvenient of current chat template implementation is that we can't distinguish between user input and special tokens (prefix/postfix)
                    const auto line_pfx = ::llama_tokenize(ctx, params.input_prefix, false, true);
-                    const auto line_inp = ::llama_tokenize(ctx, user_inp,            false, format_chat);
+                    const auto line_inp = ::llama_tokenize(ctx, user_inp,            false, params.conversation);
                    const auto line_sfx = ::llama_tokenize(ctx, params.input_suffix, false, true);

                    LOG("input tokens: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, line_inp).c_str());
@@ -31,3 +31,4 @@ for i in range(n-1):
        embedding2 = np.array(result[j])
        similarity = np.dot(embedding1, embedding2) / (np.linalg.norm(embedding1) * np.linalg.norm(embedding2))
        print(f"Similarity between {i} and {j}: {similarity:.2f}")
+
@@ -73,7 +73,6 @@ The project is under active development, and we are [looking for feedback and co
 - `-fa`, `--flash-attn` : enable flash attention (default: disabled).
 - `-ctk TYPE`, `--cache-type-k TYPE` : KV cache data type for K (default: `f16`, options `f32`, `f16`, `q8_0`, `q4_0`, `q4_1`, `iq4_nl`, `q5_0`, or `q5_1`)
 - `-ctv TYPE`, `--cache-type-v TYPE` : KV cache type for V (default `f16`, see `-ctk` for options)
- `--spm-infill` : Use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this.

 **If compiled with `LLAMA_SERVER_SSL=ON`**
 - `--ssl-key-file FNAME`: path to file a PEM-encoded SSL private key
@@ -259,7 +259,7 @@ const GRAMMAR_RANGE_LITERAL_ESCAPE_RE = /[\n\r"\]\-\\]/g;
 const GRAMMAR_LITERAL_ESCAPES = { '\r': '\\r', '\n': '\\n', '"': '\\"', '-': '\\-', ']': '\\]' };

 const NON_LITERAL_SET = new Set('|.()[]{}*+?');
-const ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = new Set('^$.[]()|{}*+?');
+const ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = new Set('[]()|{}*+?');

 export class SchemaConverter {
  constructor(options) {
@@ -751,7 +751,7 @@ export class SchemaConverter {
    const requiredProps = sortedProps.filter(k => required.has(k));
    const optionalProps = sortedProps.filter(k => !required.has(k));

-    if (additionalProperties) {
+    if (additionalProperties !== false) {
      const subName = `${name ?? ''}${name ? '-' : ''}additional`;
      const valueRule =
        additionalProperties != null && typeof additionalProperties === 'object' ? this.visit(additionalProperties, `${subName}-value`)
@@ -2020,7 +2020,6 @@ struct server_context {
                        slot.t_start_generation = 0;

                        if (slot.infill) {
-                            const bool add_bos = llama_should_add_bos_token(model);
                            bool suff_rm_leading_spc = true;
                            if (params.input_suffix.find_first_of(' ') == 0 && params.input_suffix.size() > 1) {
                                params.input_suffix.erase(0, 1);
@@ -2036,21 +2035,16 @@ struct server_context {
                            }

                            prefix_tokens.insert(prefix_tokens.begin(), llama_token_prefix(model));
-                            suffix_tokens.insert(suffix_tokens.begin(), llama_token_suffix(model));
-
-                            auto embd_inp = params.spm_infill ? suffix_tokens : prefix_tokens;
-                            auto embd_end = params.spm_infill ? prefix_tokens : suffix_tokens;
-                            if (add_bos) {
-                                embd_inp.insert(embd_inp.begin(), llama_token_bos(model));
-                            }
-                            embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end());
+                            prefix_tokens.insert(prefix_tokens.begin(), llama_token_bos(model)); // always add BOS
+                            prefix_tokens.insert(prefix_tokens.end(),   llama_token_suffix(model));
+                            prefix_tokens.insert(prefix_tokens.end(),   suffix_tokens.begin(), suffix_tokens.end());

                            const llama_token middle_token = llama_token_middle(model);
                            if (middle_token >= 0) {
-                                embd_inp.push_back(middle_token);
+                                prefix_tokens.push_back(middle_token);
                            }

-                            prompt_tokens = embd_inp;
+                            prompt_tokens = prefix_tokens;
                        } else {
                            prompt_tokens = tokenize(slot.prompt, system_prompt.empty()); // add BOS if there isn't system prompt
                        }
@@ -52,3 +52,4 @@ Feature: Passkey / Self-extend with context shift
      #| TheBloke/Llama-2-7B-GGUF        | llama-2-7b.Q2_K.gguf        | 4096        | 3   | 16384 | 512     | 4    | 512    | 500    | 300   | 1234    | 5           | 1234           |
      #| TheBloke/Mixtral-8x7B-v0.1-GGUF | mixtral-8x7b-v0.1.Q2_K.gguf | 32768       | 2   | 16384 | 512     | 4    | 512    | 500    | 100   | 0987    | 5           | 0
      # 987           |
+
@@ -1054,3 +1054,4 @@
 </body>

 </html>
+
@@ -1058,3 +1058,4 @@
 </body>

 </html>
+
@@ -34,3 +34,4 @@ fi

 #use multiple GPUs with same max compute units
 #ZES_ENABLE_SYSMAN=1 ./build/bin/llama-cli -m models/llama-2-7b.Q4_0.gguf -p "${INPUT2}" -n 400 -e -ngl 33 -s 0
+
@@ -31,3 +31,4 @@ exit /B 0
 :ERROR
 echo comomand error: %errorlevel%
 exit /B %errorlevel%
+
@@ -7,3 +7,5 @@ set INPUT2="Building a website can be done in 10 simple steps:\nStep 1:"


 .\build\bin\main.exe -m models\llama-2-7b.Q4_0.gguf -p %INPUT2% -n 400 -e -ngl 33 -s 0
+
+
@@ -20,11 +20,11 @@
    },
    "nixpkgs": {
      "locked": {
-        "lastModified": 1719506693,
-        "narHash": "sha256-C8e9S7RzshSdHB7L+v9I51af1gDM5unhJ2xO1ywxNH8=",
+        "lastModified": 1718318537,
+        "narHash": "sha256-4Zu0RYRcAY/VWuu6awwq4opuiD//ahpc2aFHg2CWqFY=",
        "owner": "NixOS",
        "repo": "nixpkgs",
-        "rev": "b2852eb9365c6de48ffb0dc2c9562591f652242a",
+        "rev": "e9ee548d90ff586a6471b4ae80ae9cfcbceb3420",
        "type": "github"
      },
      "original": {
@@ -109,7 +109,6 @@ option(GGML_LLAMAFILE                       "ggml: use ggml SGEMM"
 option(GGML_CUDA                            "ggml: use CUDA"                                  OFF)
 option(GGML_CUDA_FORCE_DMMV                 "ggml: use dmmv instead of mmvq CUDA kernels"     OFF)
 option(GGML_CUDA_FORCE_MMQ                  "ggml: use mmq kernels instead of cuBLAS"         OFF)
-option(GGML_CUDA_FORCE_CUBLAS               "ggml: always use cuBLAS instead of mmq kernels"  OFF)
 set   (GGML_CUDA_DMMV_X   "32" CACHE STRING "ggml: x stride for dmmv CUDA kernels")
 set   (GGML_CUDA_MMV_Y     "1" CACHE STRING "ggml: y block size for mmv CUDA kernels")
 option(GGML_CUDA_F16                        "ggml: use 16 bit floats for some calculations"   OFF)
@@ -120,7 +119,6 @@ set   (GGML_CUDA_PEER_MAX_BATCH_SIZE "128" CACHE STRING
 option(GGML_CUDA_NO_PEER_COPY               "ggml: do not use peer to peer copies"            OFF)
 option(GGML_CUDA_NO_VMM                     "ggml: do not try to use CUDA VMM"                OFF)
 option(GGML_CUDA_FA_ALL_QUANTS              "ggml: compile all quants for FlashAttention"     OFF)
-option(GGML_CUDA_USE_GRAPHS                 "ggml: use CUDA graphs (llama.cpp only)"          OFF)

 option(GGML_CURL                            "ggml: use libcurl to download model from an URL" OFF)
 option(GGML_HIPBLAS                         "ggml: use hipBLAS"                               OFF)
@@ -63,3 +63,4 @@ GGML_API void ggml_backend_metal_capture_next_compute(ggml_backend_t backend);
 #ifdef __cplusplus
 }
 #endif
+
@@ -295,15 +295,12 @@ if (GGML_CUDA)

        list(APPEND GGML_CDEF_PUBLIC GGML_USE_CUDA)

+        add_compile_definitions(GGML_CUDA_USE_GRAPHS)
        add_compile_definitions(GGML_CUDA_DMMV_X=${GGML_CUDA_DMMV_X})
        add_compile_definitions(GGML_CUDA_MMV_Y=${GGML_CUDA_MMV_Y})
        add_compile_definitions(K_QUANTS_PER_ITERATION=${GGML_CUDA_KQUANTS_ITER})
        add_compile_definitions(GGML_CUDA_PEER_MAX_BATCH_SIZE=${GGML_CUDA_PEER_MAX_BATCH_SIZE})

-        if (GGML_CUDA_USE_GRAPHS)
-            add_compile_definitions(GGML_CUDA_USE_GRAPHS)
-        endif()
-
        if (GGML_CUDA_FORCE_DMMV)
            add_compile_definitions(GGML_CUDA_FORCE_DMMV)
        endif()
@@ -486,11 +483,9 @@ if (GGML_SYCL)
    add_compile_options(-I./) #include DPCT

    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-narrowing")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3")
    if (GGML_SYCL_TARGET STREQUAL "NVIDIA")
        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl-targets=nvptx64-nvidia-cuda")
-        add_compile_definitions(GGML_SYCL_WARP_SIZE=32)
-    else()
-        add_compile_definitions(GGML_SYCL_WARP_SIZE=32)
    endif()

    file(GLOB   GGML_HEADERS_SYCL "ggml-sycl/*.hpp")
@@ -1168,9 +1163,7 @@ target_link_libraries(ggml PRIVATE Threads::Threads ${GGML_EXTRA_LIBS})

 find_library(MATH_LIBRARY m)
 if (MATH_LIBRARY)
-    if (NOT WIN32 OR NOT GGML_SYCL)
-        target_link_libraries(ggml PRIVATE ${MATH_LIBRARY})
-    endif()
+    target_link_libraries(ggml PRIVATE ${MATH_LIBRARY})
 endif()

 if (BUILD_SHARED_LIBS)
@@ -106,19 +106,19 @@ typedef sycl::half2 ggml_half2;
 #define QR6_K 2

 #define QI2_XXS (QK_K / (4*QR2_XXS))
-#define QR2_XXS 4
+#define QR2_XXS 8

 #define QI2_XS (QK_K / (4*QR2_XS))
-#define QR2_XS 4
+#define QR2_XS 8

 #define QI2_S (QK_K / (4*QR2_S))
-#define QR2_S 4
+#define QR2_S 8

 #define QI3_XXS (QK_K / (4*QR3_XXS))
-#define QR3_XXS 4
+#define QR3_XXS 8

 #define QI3_XS (QK_K / (4*QR3_XS))
-#define QR3_XS 4
+#define QR3_XS 8

 #define QI1_S (QK_K / (4*QR1_S))
 #define QR1_S 8
@@ -130,10 +130,10 @@ typedef sycl::half2 ggml_half2;
 #define QR4_NL 2

 #define QI4_XS (QK_K / (4*QR4_XS))
-#define QR4_XS 2
+#define QR4_XS 8

 #define QI3_S (QK_K / (4*QR3_S))
-#define QR3_S 4
+#define QR3_S 8

 #endif // GGML_COMMON_DECL_CUDA || GGML_COMMON_DECL_HIP

@@ -1882,11 +1882,6 @@ static void ggml_cuda_mul_mat(ggml_backend_cuda_context & ctx, const ggml_tensor
    bool              use_mul_mat_q =  ggml_is_quantized(src0->type)
        && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32;

-    // if mmvq is available it's a better choice than dmmv:
-#ifndef GGML_CUDA_FORCE_DMMV
-    use_dequantize_mul_mat_vec = use_dequantize_mul_mat_vec && !use_mul_mat_vec_q;
-#endif // GGML_CUDA_FORCE_DMMV
-
    bool any_gpus_with_slow_fp16 = false;

    if (split) {
@@ -1899,15 +1894,22 @@ static void ggml_cuda_mul_mat(ggml_backend_cuda_context & ctx, const ggml_tensor
            }

            const int cc            = ggml_cuda_info().devices[id].cc;
+            use_mul_mat_vec_q       = use_mul_mat_vec_q       && cc >= MIN_CC_DP4A;
            use_mul_mat_q           = use_mul_mat_q           && ggml_cuda_should_use_mmq(src0->type, cc, src1->ne[1]);
            any_gpus_with_slow_fp16 = any_gpus_with_slow_fp16 || !fast_fp16_available(cc);
        }
    } else {
        const int cc            = ggml_cuda_info().devices[ctx.device].cc;
+        use_mul_mat_vec_q       = use_mul_mat_vec_q       && cc >= MIN_CC_DP4A;
        use_mul_mat_q           = use_mul_mat_q           && ggml_cuda_should_use_mmq(src0->type, cc, src1->ne[1]);
        any_gpus_with_slow_fp16 = any_gpus_with_slow_fp16 || !fast_fp16_available(cc);
    }

+    // if mmvq is available it's a better choice than dmmv:
+#ifndef GGML_CUDA_FORCE_DMMV
+    use_dequantize_mul_mat_vec = use_dequantize_mul_mat_vec && !use_mul_mat_vec_q;
+#endif // GGML_CUDA_FORCE_DMMV
+
    // debug helpers
    //printf("src0: %8d %8d %8d %8d\n", src0->ne[0], src0->ne[1], src0->ne[2], src0->ne[3]);
    //printf("      %8d %8d %8d %8d\n", src0->nb[0], src0->nb[1], src0->nb[2], src0->nb[3]);
@@ -2711,40 +2713,27 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
        case GGML_OP_MUL_MAT:
        case GGML_OP_MUL_MAT_ID:
            {
-                struct ggml_tensor * a = op->src[0];
+                struct ggml_tensor * a;
+                struct ggml_tensor * b;
                if (op->op == GGML_OP_MUL_MAT) {
-                    struct ggml_tensor * b = op->src[1];
-                    if (a->ne[3] != b->ne[3]) {
+                    a = op->src[0];
+                    b = op->src[1];
+                } else {
+                    a = op->src[2];
+                    b = op->src[1];
+                }
+                if (a->ne[3] != b->ne[3]) {
+                    return false;
+                }
+                ggml_type a_type = a->type;
+                if (a_type == GGML_TYPE_IQ2_XXS || a_type == GGML_TYPE_IQ2_XS || a_type == GGML_TYPE_IQ3_XXS ||
+                    a_type == GGML_TYPE_IQ1_S   || a_type == GGML_TYPE_IQ4_NL || a_type == GGML_TYPE_IQ3_S   ||
+                    a_type == GGML_TYPE_IQ1_M   || a_type == GGML_TYPE_IQ2_S  || a_type == GGML_TYPE_IQ4_XS) {
+                    if (b->ne[1] == 1 && ggml_nrows(b) > 1) {
                        return false;
                    }
                }
-                switch (a->type) {
-                    case GGML_TYPE_F32:
-                    case GGML_TYPE_F16:
-                    case GGML_TYPE_Q4_0:
-                    case GGML_TYPE_Q4_1:
-                    case GGML_TYPE_Q5_0:
-                    case GGML_TYPE_Q5_1:
-                    case GGML_TYPE_Q8_0:
-                    case GGML_TYPE_Q2_K:
-                    case GGML_TYPE_Q3_K:
-                    case GGML_TYPE_Q4_K:
-                    case GGML_TYPE_Q5_K:
-                    case GGML_TYPE_Q6_K:
-                    case GGML_TYPE_Q8_K:
-                    case GGML_TYPE_IQ1_M:
-                    case GGML_TYPE_IQ1_S:
-                    case GGML_TYPE_IQ2_S:
-                    case GGML_TYPE_IQ2_XS:
-                    case GGML_TYPE_IQ2_XXS:
-                    case GGML_TYPE_IQ3_S:
-                    case GGML_TYPE_IQ3_XXS:
-                    case GGML_TYPE_IQ4_NL:
-                    case GGML_TYPE_IQ4_XS:
-                        return true;
-                    default:
-                        return false;
-                }
+                return true;
            } break;
        case GGML_OP_GET_ROWS:
            {
@@ -3,7 +3,6 @@
 #include "ggml.h"
 #include "ggml-cuda.h"

-#include <cstdint>
 #include <memory>

 #if defined(GGML_USE_HIPBLAS)
@@ -269,15 +268,30 @@ static __device__ __forceinline__ unsigned int __vcmpeq4(unsigned int a, unsigne
    return c;
 }

-static __device__ __forceinline__ unsigned int __vcmpne4(unsigned int a, unsigned int b) {
-    const uint8x4_t& va = reinterpret_cast<const uint8x4_t&>(a);
-    const uint8x4_t& vb = reinterpret_cast<const uint8x4_t&>(b);
-    unsigned int c;
-    uint8x4_t& vc = reinterpret_cast<uint8x4_t&>(c);
-#pragma unroll
-    for (int i = 0; i < 4; ++i) {
-        vc[i] = va[i] == vb[i] ? 0x00 : 0xff;
-    }
+static __device__ __forceinline__ int __dp4a(const int a, const int b, int c) {
+#if defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx1030__)
+    c = __builtin_amdgcn_sdot4(a, b, c, false);
+#elif defined(RDNA3)
+    c = __builtin_amdgcn_sudot4( true, a, true, b, c, false);
+#elif defined(__gfx1010__) || defined(__gfx900__)
+    int tmp1;
+    int tmp2;
+    asm("\n \
+        v_mul_i32_i24 %1, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0 \n \
+        v_mul_i32_i24 %2, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1 \n \
+        v_add3_u32 %0, %1, %2, %0 \n \
+        v_mul_i32_i24 %1, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_2 src1_sel:BYTE_2 \n \
+        v_mul_i32_i24 %2, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:BYTE_3 \n \
+        v_add3_u32 %0, %1, %2, %0 \n \
+        "
+        : "+v"(c), "=&v"(tmp1), "=&v"(tmp2)
+        : "v"(a), "v"(b)
+    );
+#else
+    const int8x4_t va = reinterpret_cast<const int8x4_t&>(a);
+    const int8x4_t vb = reinterpret_cast<const int8x4_t&>(b);
+    c += va[0] * vb[0] + va[1] * vb[1] + va[2] * vb[2] + va[3] * vb[3];
+#endif
    return c;
 }

@@ -453,48 +467,8 @@ static __device__ __forceinline__ uint32_t __hgt2_mask(const half2 a, const half
 }
 #endif // CUDART_VERSION < 12000

-static __device__ __forceinline__ int ggml_cuda_dp4a(const int a, const int b, int c) {
-#if defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)
-#if defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx1030__)
-    c = __builtin_amdgcn_sdot4(a, b, c, false);
-#elif defined(RDNA3)
-    c = __builtin_amdgcn_sudot4( true, a, true, b, c, false);
-#elif defined(__gfx1010__) || defined(__gfx900__)
-    int tmp1;
-    int tmp2;
-    asm("\n \
-        v_mul_i32_i24 %1, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0 \n \
-        v_mul_i32_i24 %2, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1 \n \
-        v_add3_u32 %0, %1, %2, %0 \n \
-        v_mul_i32_i24 %1, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_2 src1_sel:BYTE_2 \n \
-        v_mul_i32_i24 %2, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:BYTE_3 \n \
-        v_add3_u32 %0, %1, %2, %0 \n \
-        "
-        : "+v"(c), "=&v"(tmp1), "=&v"(tmp2)
-        : "v"(a), "v"(b)
-    );
-#else
-    const int8x4_t va = reinterpret_cast<const int8x4_t&>(a);
-    const int8x4_t vb = reinterpret_cast<const int8x4_t&>(b);
-    c += va[0] * vb[0] + va[1] * vb[1] + va[2] * vb[2] + va[3] * vb[3];
-#endif
-    return c;
-
-#else // defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)
-
-#if __CUDA_ARCH__ >= MIN_CC_DP4A
-    return __dp4a(a, b, c);
-#else // __CUDA_ARCH__ >= MIN_CC_DP4A
-    const int8_t * a8 = (const int8_t *) &a;
-    const int8_t * b8 = (const int8_t *) &b;
-    return c + a8[0]*b8[0] + a8[1]*b8[1] + a8[2]*b8[2] + a8[3]*b8[3];
-#endif // __CUDA_ARCH__ >= MIN_CC_DP4A
-
-#endif // defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)
-}
-
 // TODO: move to ggml-common.h
-static constexpr __device__ int8_t kvalues_iq4nl[16] = {-127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113};
+static const __device__ int8_t kvalues_iq4nl[16] = {-127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113};

 typedef void (*dequantize_kernel_t)(const void * vx, const int64_t ib, const int iqs, dfloat2 & v);

@@ -487,3 +487,4 @@ void* ggml_cuda_cpy_fn(const ggml_tensor * src0, ggml_tensor * src1) {
        GGML_ASSERT(false);
    }
 }
+
@@ -54,11 +54,12 @@ typedef float (*vec_dot_KQ_f32_t)(
 template<typename T, int D>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_0(
    const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
+#if __CUDA_ARCH__ >= MIN_CC_DP4A

    const block_q4_0 * K_q4_0 = (const block_q4_0 *) K_c;
    GGML_UNUSED(Q_v);

-    T sum = 0.0f;
+    half sum = 0.0f;

 #pragma unroll
    for (int k_KQ_0 = 0; k_KQ_0 < D/sizeof(int); k_KQ_0 += WARP_SIZE) {
@@ -71,7 +72,7 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_0(
        const int v = (get_int_from_uint8(K_q4_0[ib].qs, iqs4) >> shift) & 0x0F0F0F0F;
        const int u = Q_q8[k_KQ_0/WARP_SIZE];

-        const int sumi = ggml_cuda_dp4a(v, u, 0);
+        const int sumi = __dp4a(v, u, 0);

 #ifdef FP16_AVAILABLE
        if (std::is_same<T, half>::value) {
@@ -89,11 +90,19 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_0(
    }

    return sum;
+#else
+    GGML_UNUSED(K_c);
+    GGML_UNUSED(Q_v);
+    GGML_UNUSED(Q_q8);
+    GGML_UNUSED(Q_ds_v);
+    NO_DEVICE_CODE;
+#endif  // __CUDA_ARCH__ >= MIN_CC_DP4A
 }

 template<typename T, int D>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_1(
    const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
+#if __CUDA_ARCH__ >= MIN_CC_DP4A

    const block_q4_1 * K_q4_1 = (const block_q4_1 *) K_c;
    GGML_UNUSED(Q_v);
@@ -111,7 +120,7 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_1(
        const int v = (get_int_from_uint8_aligned(K_q4_1[ib].qs, iqs4) >> shift) & 0x0F0F0F0F;
        const int u = Q_q8[k_KQ_0/WARP_SIZE];

-        const int sumi = ggml_cuda_dp4a(v, u, 0);
+        const int sumi = __dp4a(v, u, 0);

 #ifdef FP16_AVAILABLE
        if (std::is_same<T, half>::value) {
@@ -133,11 +142,19 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_1(
    }

    return sum;
+#else
+    GGML_UNUSED(K_c);
+    GGML_UNUSED(Q_v);
+    GGML_UNUSED(Q_q8);
+    GGML_UNUSED(Q_ds_v);
+    NO_DEVICE_CODE;
+#endif  // __CUDA_ARCH__ >= MIN_CC_DP4A
 }

 template<typename T, int D>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_0(
    const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
+#if __CUDA_ARCH__ >= MIN_CC_DP4A

    const block_q5_0 * K_q5_0 = (const block_q5_0 *) K_c;
    GGML_UNUSED(Q_v);
@@ -162,7 +179,7 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_0(

        const int u = Q_q8[k_KQ_0/WARP_SIZE];

-        const int sumi = ggml_cuda_dp4a(v, u, 0);
+        const int sumi = __dp4a(v, u, 0);

 #ifdef FP16_AVAILABLE
        if (std::is_same<T, half>::value) {
@@ -180,11 +197,19 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_0(
    }

    return sum;
+#else
+    GGML_UNUSED(K_c);
+    GGML_UNUSED(Q_v);
+    GGML_UNUSED(Q_q8);
+    GGML_UNUSED(Q_ds_v);
+    NO_DEVICE_CODE;
+#endif  // __CUDA_ARCH__ >= MIN_CC_DP4A
 }

 template<typename T, int D>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_1(
    const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
+#if __CUDA_ARCH__ >= MIN_CC_DP4A

    const block_q5_1 * K_q5_1 = (const block_q5_1 *) K_c;
    GGML_UNUSED(Q_v);
@@ -209,7 +234,7 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_1(

        const int u = Q_q8[k_KQ_0/WARP_SIZE];

-        const int sumi = ggml_cuda_dp4a(v, u, 0);
+        const int sumi = __dp4a(v, u, 0);

 #ifdef FP16_AVAILABLE
        if (std::is_same<T, half>::value) {
@@ -231,11 +256,19 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_1(
    }

    return sum;
+#else
+    GGML_UNUSED(K_c);
+    GGML_UNUSED(Q_v);
+    GGML_UNUSED(Q_q8);
+    GGML_UNUSED(Q_ds_v);
+    NO_DEVICE_CODE;
+#endif  // __CUDA_ARCH__ >= MIN_CC_DP4A
 }

 template <typename T, int D>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q8_0(
    const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
+#if __CUDA_ARCH__ >= MIN_CC_DP4A

    const block_q8_0 * K_q8_0 = (const block_q8_0 *) K_c;
    GGML_UNUSED(Q_v);
@@ -264,6 +297,13 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q8_0(
    }

    return sum;
+#else
+    GGML_UNUSED(K_c);
+    GGML_UNUSED(Q_v);
+    GGML_UNUSED(Q_q8);
+    GGML_UNUSED(Q_ds_v);
+    NO_DEVICE_CODE;
+#endif  // __CUDA_ARCH__ >= MIN_CC_DP4A
 }

 template <typename T, int D>
@@ -2475,7 +2475,7 @@ static void launch_mul_mat_q(ggml_backend_cuda_context & ctx, const mmq_args & a

    const dim3 block_nums_mmq(nsm, 1, 1);

-    ggml_cuda_pool & pool = ctx.pool(id);
+    ggml_cuda_pool & pool = ctx.pool();
    ggml_cuda_pool_alloc<float> tmp_fixup(pool, block_nums_mmq.x * mmq_x*mmq_y);

    if (args.ne01 % mmq_y == 0) {
@@ -28,22 +28,16 @@ static constexpr __device__ vec_dot_q_cuda_t get_vec_dot_q_cuda(ggml_type type)

 static constexpr __device__ int get_vdr_mmvq(ggml_type type) {
    return type == GGML_TYPE_Q4_0 ? VDR_Q4_0_Q8_1_MMVQ :
-        type == GGML_TYPE_Q4_1    ? VDR_Q4_1_Q8_1_MMVQ :
-        type == GGML_TYPE_Q5_0    ? VDR_Q5_0_Q8_1_MMVQ :
-        type == GGML_TYPE_Q5_1    ? VDR_Q5_1_Q8_1_MMVQ :
-        type == GGML_TYPE_Q8_0    ? VDR_Q8_0_Q8_1_MMVQ :
-        type == GGML_TYPE_Q2_K    ? VDR_Q2_K_Q8_1_MMVQ :
-        type == GGML_TYPE_Q3_K    ? VDR_Q3_K_Q8_1_MMVQ :
-        type == GGML_TYPE_Q4_K    ? VDR_Q4_K_Q8_1_MMVQ :
-        type == GGML_TYPE_Q5_K    ? VDR_Q5_K_Q8_1_MMVQ :
-        type == GGML_TYPE_Q6_K    ? VDR_Q6_K_Q8_1_MMVQ :
-        type == GGML_TYPE_IQ2_XXS ? VDR_IQ2_XXS_Q8_1_MMVQ :
-        type == GGML_TYPE_IQ2_XS  ? VDR_IQ2_XS_Q8_1_MMVQ :
-        type == GGML_TYPE_IQ2_S   ? VDR_IQ2_S_Q8_1_MMVQ :
-        type == GGML_TYPE_IQ3_XXS ? VDR_IQ3_XXS_Q8_1_MMVQ :
-        type == GGML_TYPE_IQ3_S   ? VDR_IQ3_S_Q8_1_MMVQ :
-        type == GGML_TYPE_IQ4_NL  ? VDR_IQ4_NL_Q8_1_MMVQ :
-        type == GGML_TYPE_IQ4_XS  ? VDR_IQ4_XS_Q8_1_MMVQ :
+        type == GGML_TYPE_Q4_1 ? VDR_Q4_1_Q8_1_MMVQ :
+        type == GGML_TYPE_Q5_0 ? VDR_Q5_0_Q8_1_MMVQ :
+        type == GGML_TYPE_Q5_1 ? VDR_Q5_1_Q8_1_MMVQ :
+        type == GGML_TYPE_Q8_0 ? VDR_Q8_0_Q8_1_MMVQ :
+        type == GGML_TYPE_Q2_K ? VDR_Q2_K_Q8_1_MMVQ :
+        type == GGML_TYPE_Q3_K ? VDR_Q3_K_Q8_1_MMVQ :
+        type == GGML_TYPE_Q4_K ? VDR_Q4_K_Q8_1_MMVQ :
+        type == GGML_TYPE_Q5_K ? VDR_Q5_K_Q8_1_MMVQ :
+        type == GGML_TYPE_Q6_K ? VDR_Q6_K_Q8_1_MMVQ :
+        type == GGML_TYPE_IQ4_NL ? VDR_Q4_K_Q8_1_MMVQ :
        1;
 }

@@ -6537,3 +6537,4 @@ template [[host_name("kernel_mul_mv_id_iq3_s_f32")]]   kernel kernel_mul_mv_id_t
 template [[host_name("kernel_mul_mv_id_iq2_s_f32")]]   kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq2_s_f32_impl>>;
 template [[host_name("kernel_mul_mv_id_iq4_nl_f32")]]  kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq4_nl_f32_impl>>;
 template [[host_name("kernel_mul_mv_id_iq4_xs_f32")]]  kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq4_xs_f32_impl>>;
+
@@ -130,3 +130,4 @@ void iq3xs_free_impl(int grid_size);
 #ifdef __cplusplus
 }
 #endif
+
@@ -74,6 +74,51 @@ typedef void (*ggml_sycl_op_flatten_t)(ggml_backend_sycl_context & ctx, const gg
                                       const float *src1_dd, float *dst_dd,
                                       const queue_ptr &main_stream);

+static __dpct_inline__ float warp_reduce_sum(float x,
+                                             const sycl::nd_item<3> &item_ct1) {
+#pragma unroll
+    for (int mask = 16; mask > 0; mask >>= 1) {
+        /*
+        DPCT1096:98: The right-most dimension of the work-group used in the SYCL
+        kernel that calls this function may be less than "32". The function
+        "dpct::permute_sub_group_by_xor" may return an unexpected result on the
+        CPU device. Modify the size of the work-group to ensure that the value
+        of the right-most dimension is a multiple of "32".
+        */
+        x += dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), x, mask);
+    }
+    return x;
+}
+
+static __dpct_inline__ sycl::float2
+warp_reduce_sum(sycl::float2 a, const sycl::nd_item<3> &item_ct1) {
+#pragma unroll
+    for (int mask = 16; mask > 0; mask >>= 1) {
+        a.x() += dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), a.x(),
+                                                mask);
+        a.y() += dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), a.y(),
+                                                mask);
+    }
+    return a;
+}
+
+static __dpct_inline__ float warp_reduce_max(float x,
+                                             const sycl::nd_item<3> &item_ct1) {
+#pragma unroll
+    for (int mask = 16; mask > 0; mask >>= 1) {
+        /*
+        DPCT1096:97: The right-most dimension of the work-group used in the SYCL
+        kernel that calls this function may be less than "32". The function
+        "dpct::permute_sub_group_by_xor" may return an unexpected result on the
+        CPU device. Modify the size of the work-group to ensure that the value
+        of the right-most dimension is a multiple of "32".
+        */
+        x = sycl::fmax(x, dpct::permute_sub_group_by_xor(
+                              item_ct1.get_sub_group(), x, mask));
+    }
+    return x;
+}
+
 static __dpct_inline__ float op_repeat(const float a, const float b) {
    return b;
    GGML_UNUSED(a);
@@ -291,6 +336,47 @@ static void sqr_f32(const float * x, float * dst, const int k,
    dst[i] = x[i] * x[i];
 }

+static void norm_f32(const float * x, float * dst, const int ncols, const float eps,
+                     const sycl::nd_item<3> &item_ct1, sycl::float2 *s_sum, int block_size) {
+    const int row = item_ct1.get_group(2) * item_ct1.get_local_range(1) +
+                    item_ct1.get_local_id(1);
+    const int tid = item_ct1.get_local_id(2);
+
+    sycl::float2 mean_var = sycl::float2(0.f, 0.f);
+
+    for (int col = tid; col < ncols; col += block_size) {
+        const float xi = x[row*ncols + col];
+        mean_var.x() += xi;
+        mean_var.y() += xi * xi;
+    }
+
+    // sum up partial sums
+    mean_var = warp_reduce_sum(mean_var, item_ct1);
+    if (block_size > WARP_SIZE) {
+
+        int warp_id = item_ct1.get_local_id(2) / WARP_SIZE;
+        int lane_id = item_ct1.get_local_id(2) % WARP_SIZE;
+        if (lane_id == 0) {
+            s_sum[warp_id] = mean_var;
+        }
+        /*
+        DPCT1118:0: SYCL group functions and algorithms must be encountered in
+        converged control flow. You may need to adjust the code.
+        */
+        item_ct1.barrier(sycl::access::fence_space::local_space);
+        mean_var = s_sum[lane_id];
+        mean_var = warp_reduce_sum(mean_var, item_ct1);
+    }
+
+    const float mean = mean_var.x() / ncols;
+    const float var = mean_var.y() / ncols - mean * mean;
+    const float inv_std = sycl::rsqrt(var + eps);
+
+    for (int col = tid; col < ncols; col += block_size) {
+        dst[row*ncols + col] = (x[row*ncols + col] - mean) * inv_std;
+    }
+}
+
 static void concat_f32(const float  *x,const float  *y, float *dst, const int ne0, const int ne02,
                       const sycl::nd_item<3> &item_ct1) {
    int nidx = item_ct1.get_local_id(2) +
@@ -358,11 +444,126 @@ static void pad_f32(const float  *x, float *dst, const int ne0, const int ne00,
    }
 }

-template<int QUANT_BLOCK_TILE>
+static void group_norm_f32(const float * x, float * dst, const int group_size, const int ne_elements, const float eps,
+                           const sycl::nd_item<3> &item_ct1, float *s_sum, int block_size) {
+    int start = item_ct1.get_group(2) * group_size;
+    int end = start + group_size;
+
+    start += item_ct1.get_local_id(2);
+
+    if (end >= ne_elements) {
+        end = ne_elements;
+    }
+
+    float tmp = 0.0f; // partial sum for thread in warp
+
+    for (int j = start; j < end; j += block_size) {
+        tmp += x[j];
+    }
+
+    tmp = warp_reduce_sum(tmp, item_ct1);
+    if (block_size > WARP_SIZE) {
+
+        int warp_id = item_ct1.get_local_id(2) / WARP_SIZE;
+        int lane_id = item_ct1.get_local_id(2) % WARP_SIZE;
+        if (lane_id == 0) {
+            s_sum[warp_id] = tmp;
+        }
+        /*
+        DPCT1118:1: SYCL group functions and algorithms must be encountered in
+        converged control flow. You may need to adjust the code.
+        */
+        /*
+        DPCT1065:54: Consider replacing sycl::nd_item::barrier() with
+        sycl::nd_item::barrier(sycl::access::fence_space::local_space) for
+        better performance if there is no access to global memory.
+        */
+        item_ct1.barrier();
+        tmp = s_sum[lane_id];
+        tmp = warp_reduce_sum(tmp, item_ct1);
+    }
+
+    float mean = tmp / group_size;
+    tmp = 0.0f;
+
+    for (int j = start; j < end; j += block_size) {
+        float xi = x[j] - mean;
+        dst[j] = xi;
+        tmp += xi * xi;
+    }
+
+    tmp = warp_reduce_sum(tmp, item_ct1);
+    if (block_size > WARP_SIZE) {
+
+        int warp_id = item_ct1.get_local_id(2) / WARP_SIZE;
+        int lane_id = item_ct1.get_local_id(2) % WARP_SIZE;
+        if (lane_id == 0) {
+            s_sum[warp_id] = tmp;
+        }
+        /*
+        DPCT1118:2: SYCL group functions and algorithms must be encountered in
+        converged control flow. You may need to adjust the code.
+        */
+        /*
+        DPCT1065:55: Consider replacing sycl::nd_item::barrier() with
+        sycl::nd_item::barrier(sycl::access::fence_space::local_space) for
+        better performance if there is no access to global memory.
+        */
+        item_ct1.barrier();
+        tmp = s_sum[lane_id];
+        tmp = warp_reduce_sum(tmp, item_ct1);
+    }
+
+    float variance = tmp / group_size;
+    float scale = sycl::rsqrt(variance + eps);
+    for (int j = start; j < end; j += block_size) {
+        dst[j] *= scale;
+    }
+}
+
+static void rms_norm_f32(const float * x, float * dst, const int ncols, const float eps,
+                         const sycl::nd_item<3> &item_ct1, float *s_sum, int block_size) {
+    const int row = item_ct1.get_group(2) * item_ct1.get_local_range(1) +
+                    item_ct1.get_local_id(1);
+    const int tid = item_ct1.get_local_id(2);
+
+    float tmp = 0.0f; // partial sum for thread in warp
+
+    for (int col = tid; col < ncols; col += block_size) {
+        const float xi = x[row*ncols + col];
+        tmp += xi * xi;
+    }
+
+    // sum up partial sums
+    tmp = warp_reduce_sum(tmp, item_ct1);
+    if (block_size > WARP_SIZE) {
+
+        int warp_id = item_ct1.get_local_id(2) / WARP_SIZE;
+        int lane_id = item_ct1.get_local_id(2) % WARP_SIZE;
+        if (lane_id == 0) {
+            s_sum[warp_id] = tmp;
+        }
+        /*
+        DPCT1118:3: SYCL group functions and algorithms must be encountered in
+        converged control flow. You may need to adjust the code.
+        */
+        item_ct1.barrier(sycl::access::fence_space::local_space);
+        tmp = s_sum[lane_id];
+        tmp = warp_reduce_sum(tmp, item_ct1);
+    }
+
+    const float mean = tmp / ncols;
+    const float scale = sycl::rsqrt(mean + eps);
+
+    for (int col = tid; col < ncols; col += block_size) {
+        dst[row*ncols + col] = scale * x[row*ncols + col];
+    }
+}
+
 static void quantize_q8_1(const float * __restrict__ x, void * __restrict__ vy, const int kx, const int kx_padded,
                          const sycl::nd_item<3> &item_ct1) {
-    const int ix = (item_ct1.get_local_range(2) * item_ct1.get_group(2) +
-                    item_ct1.get_local_id(2)) * QUANT_BLOCK_TILE;
+    const int ix = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
+                   item_ct1.get_local_id(2);

    if (ix >= kx_padded) {
        return;
@@ -377,39 +578,23 @@ static void quantize_q8_1(const float * __restrict__ x, void * __restrict__ vy,

    const int ib = i_padded / QK8_1; // block index
    const int iqs = i_padded % QK8_1; // quant index
-    typedef  sycl::vec<float, QUANT_BLOCK_TILE> TC;
-    typedef  sycl::vec<int8_t, QUANT_BLOCK_TILE> TQ;
-    TC zeros;
-    TQ qzeros;
+
+    const float xi = ix < kx ? x[iy*kx + ix] : 0.0f;
+    float amax = sycl::fabs((float)xi);
+    float sum = xi;
+
 #pragma unroll
-    for (int i = 0; i < QUANT_BLOCK_TILE; i++)
-    {
-        zeros[i] = 0.f;
-        qzeros[i] = 0;
+    for (int mask = 16; mask > 0; mask >>= 1) {
+        amax = sycl::fmax(amax, dpct::permute_sub_group_by_xor(
+                                    item_ct1.get_sub_group(), amax, mask));
+        sum +=
+            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), sum, mask);
    }
-    const TC xi = ix < kx ? *(TC *)&x[iy * kx + ix] : zeros;
-    float sum = xi[0];
-    float amax = sycl::fabs(xi[0]);
-#pragma unroll
-    for (int i = 1; i < QUANT_BLOCK_TILE; i++)
-    {
-        sum += xi[i];
-        amax = sycl::fmax(sycl::fabs(xi[i]), amax);
-    }
-    sum = warp_reduce_sum(sum, item_ct1);
-    amax = warp_reduce_max(amax, item_ct1);

    const float d = amax / 127;
-    TQ q = qzeros;
-    if (amax != 0.0f)
-    {
-#pragma unroll
-        for (int i = 0; i < QUANT_BLOCK_TILE; i++) {
-            q[i] = sycl::round(xi[i] / d);
-        }
-    }
+    const int8_t q = amax == 0.0f ? 0 : sycl::round(xi / d);

-    *(TQ *)&y[ib].qs[iqs] = q;
+    y[ib].qs[iqs] = q;

    if (iqs > 0) {
        return;
@@ -543,7 +728,7 @@ static void mul_mat_p021_f16_f32(

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -596,7 +781,7 @@ static void mul_mat_vec_nc_f16_f32( // nc == non-contiguous

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -793,6 +978,114 @@ static void cpy_f32_q(const char * cx, char * cdst, const int ne,
    cpy_blck(cx + x_offset, cdst + dst_offset);
 }

+static float rope_yarn_ramp(const float low, const float high, const int i0) {
+    const float y = (i0 / 2 - low) / sycl::max(0.001f, high - low);
+    return 1.0f - sycl::min(1.0f, sycl::max(0.0f, y));
+}
+
+struct rope_corr_dims {
+    float v[4];
+};
+
+// YaRN algorithm based on LlamaYaRNScaledRotaryEmbedding.py from https://github.com/jquesnelle/yarn
+// MIT licensed. Copyright (c) 2023 Jeffrey Quesnelle and Bowen Peng.
+static void rope_yarn(
+    float theta_extrap, float freq_scale, rope_corr_dims corr_dims, int64_t i0, float ext_factor, float mscale,
+    float * cos_theta, float * sin_theta
+) {
+    // Get n-d rotational scaling corrected for extrapolation
+    float theta_interp = freq_scale * theta_extrap;
+    float theta = theta_interp;
+    if (ext_factor != 0.0f) {
+        float ramp_mix = rope_yarn_ramp(corr_dims.v[0], corr_dims.v[1], i0) * ext_factor;
+        theta = theta_interp * (1 - ramp_mix) + theta_extrap * ramp_mix;
+
+        // Get n-d magnitude scaling corrected for interpolation
+        mscale *= 1.0f + 0.1f * sycl::log(1.0f / freq_scale);
+    }
+    *cos_theta = sycl::cos(theta) * mscale;
+    *sin_theta = sycl::sin(theta) * mscale;
+}
+
+// rope == RoPE == rotary positional embedding
+template<typename T, bool has_pos>
+static void rope(
+    const T * x, T * dst, int ncols, const int32_t * pos, float freq_scale, int p_delta_rows, float freq_base,
+    float ext_factor, float attn_factor, rope_corr_dims corr_dims
+,
+    const sycl::nd_item<3> &item_ct1) {
+    const int col = 2 * (item_ct1.get_local_range(1) * item_ct1.get_group(1) +
+                         item_ct1.get_local_id(1));
+
+    if (col >= ncols) {
+        return;
+    }
+
+    const int row = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
+                    item_ct1.get_local_id(2);
+    const int i = row*ncols + col;
+    const int i2 = row/p_delta_rows;
+
+    const int p = has_pos ? pos[i2] : 0;
+    const float theta_base = p * dpct::pow(freq_base, -float(col) / ncols);
+
+    float cos_theta, sin_theta;
+    rope_yarn(theta_base, freq_scale, corr_dims, col, ext_factor, attn_factor, &cos_theta, &sin_theta);
+
+    const float x0 = x[i + 0];
+    const float x1 = x[i + 1];
+
+    dst[i + 0] = x0*cos_theta - x1*sin_theta;
+    dst[i + 1] = x0*sin_theta + x1*cos_theta;
+}
+
+template<typename T, bool has_pos, bool has_freq_facs>
+static void rope_neox(
+    const T * x, T * dst, int ncols, int n_dims, const int32_t * pos, float freq_scale, int p_delta_rows,
+    float ext_factor, float attn_factor, rope_corr_dims corr_dims, float theta_scale, float inv_ndims,
+    const float * freq_factors, const sycl::nd_item<3> &item_ct1) {
+    const int col = 2 * (item_ct1.get_local_range(1) * item_ct1.get_group(1) +
+                         item_ct1.get_local_id(1));
+
+    if (col >= ncols) {
+        return;
+    }
+
+    const int row = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
+                    item_ct1.get_local_id(2);
+    const int ib = col / n_dims;
+    const int ic = col % n_dims;
+
+    if (ib > 0) {
+        const int i = row*ncols + ib*n_dims + ic;
+
+        dst[i + 0] = x[i + 0];
+        dst[i + 1] = x[i + 1];
+
+        return;
+    }
+
+    const int i  = row*ncols + ib*n_dims + ic/2;
+    const int i2 = row/p_delta_rows;
+
+    float cur_rot = inv_ndims * ic - ib;
+
+    const int p = has_pos ? pos[i2] : 0;
+    const float freq_factor = has_freq_facs ? freq_factors[ic/2] : 1.0f;
+
+    const float theta_base =
+        p * freq_scale * dpct::pow(theta_scale, col / 2.0f)/freq_factor;
+
+    float cos_theta, sin_theta;
+    rope_yarn(theta_base, freq_scale, corr_dims, cur_rot, ext_factor, attn_factor, &cos_theta, &sin_theta);
+
+    const float x0 = x[i + 0];
+    const float x1 = x[i + n_dims/2];
+
+    dst[i + 0]        = x0*cos_theta - x1*sin_theta;
+    dst[i + n_dims/2] = x0*sin_theta + x1*cos_theta;
+}
+
 static void k_sum_rows_f32(const float * x, float * dst, const int ncols,
                           const sycl::nd_item<3> &item_ct1) {
    const int row = item_ct1.get_group(1);
@@ -1458,6 +1751,99 @@ static void sqr_f32_sycl(const float *x, float *dst, const int k,
        });
 }

+static void norm_f32_sycl(const float *x, float *dst, const int ncols,
+                          const int nrows, const float eps,
+                          queue_ptr stream) {
+    GGML_ASSERT(ncols % WARP_SIZE == 0);
+    if (ncols < 1024) {
+        const sycl::range<3> block_dims(1, 1, WARP_SIZE);
+        stream->submit([&](sycl::handler &cgh) {
+            sycl::local_accessor<sycl::float2, 1> s_sum_acc_ct1(
+                sycl::range<1>(32), cgh);
+
+            cgh.parallel_for(
+                sycl::nd_range<3>(sycl::range<3>(1, 1, nrows) * block_dims,
+                                  block_dims),
+                [=](sycl::nd_item<3> item_ct1)
+                    [[intel::reqd_sub_group_size(32)]] {
+                        norm_f32(x, dst, ncols, eps, item_ct1,
+                                            s_sum_acc_ct1.get_pointer(), WARP_SIZE);
+                    });
+        });
+    } else {
+        const int work_group_size = get_work_group_size(stream->get_device());
+        const sycl::range<3> block_dims(1, 1, work_group_size);
+        /*
+        DPCT1049:17: The work-group size passed to the SYCL kernel may exceed
+        the limit. To get the device limit, query
+        info::device::max_work_group_size. Adjust the work-group size if needed.
+        */
+        stream->submit([&](sycl::handler &cgh) {
+            sycl::local_accessor<sycl::float2, 1> s_sum_acc_ct1(
+                sycl::range<1>(32), cgh);
+
+            cgh.parallel_for(
+                sycl::nd_range<3>(sycl::range<3>(1, 1, nrows) * block_dims,
+                                  block_dims),
+                [=](sycl::nd_item<3> item_ct1)
+                    [[intel::reqd_sub_group_size(32)]] {
+                        norm_f32(x, dst, ncols, eps, item_ct1,
+                                       s_sum_acc_ct1.get_pointer(), work_group_size);
+                    });
+        });
+    }
+}
+
+static void group_norm_f32_sycl(const float *x, float *dst,
+                                const int num_groups, const int group_size,
+                                const int ne_elements, queue_ptr stream) {
+    static const float eps = 1e-6f;
+    if (group_size < 1024) {
+        const sycl::range<3> block_dims(1, 1, WARP_SIZE);
+        stream->submit([&](sycl::handler &cgh) {
+            sycl::local_accessor<float, 1> s_sum_acc_ct1(sycl::range<1>(32),
+                                                         cgh);
+
+            const float eps_ct4 = eps;
+
+            cgh.parallel_for(
+                sycl::nd_range<3>(sycl::range<3>(1, 1, num_groups) * block_dims,
+                                  block_dims),
+                [=](sycl::nd_item<3> item_ct1)
+                    [[intel::reqd_sub_group_size(32)]] {
+                        group_norm_f32(
+                            x, dst, group_size, ne_elements, eps_ct4, item_ct1,
+                            s_sum_acc_ct1.get_pointer(), WARP_SIZE);
+                    });
+        });
+    } else {
+        const int work_group_size = get_work_group_size(stream->get_device());
+        const sycl::range<3> block_dims(1, 1, work_group_size);
+        /*
+        DPCT1049:18: The work-group size passed to the SYCL kernel may exceed
+        the limit. To get the device limit, query
+        info::device::max_work_group_size. Adjust the work-group size if needed.
+        */
+
+        stream->submit([&](sycl::handler &cgh) {
+            sycl::local_accessor<float, 1> s_sum_acc_ct1(sycl::range<1>(32),
+                                                         cgh);
+
+            const float eps_ct4 = eps;
+
+            cgh.parallel_for(
+                sycl::nd_range<3>(sycl::range<3>(1, 1, num_groups) * block_dims,
+                                  block_dims),
+                [=](sycl::nd_item<3> item_ct1)
+                    [[intel::reqd_sub_group_size(32)]] {
+                        group_norm_f32(x, dst, group_size, ne_elements,
+                                             eps_ct4, item_ct1,
+                                             s_sum_acc_ct1.get_pointer(), work_group_size);
+                    });
+        });
+    }
+}
+
 static void concat_f32_sycl(const float *x, const float *y, float *dst,
                            const int ne0, int ne1, int ne2, int ne02,
                            queue_ptr stream) {
@@ -1499,22 +1885,64 @@ static void pad_f32_sycl(const float *x, float *dst, const int ne00,
        });
 }

+static void rms_norm_f32_sycl(const float *x, float *dst, const int ncols,
+                              const int nrows, const float eps,
+                              queue_ptr stream) {
+    GGML_ASSERT(ncols % WARP_SIZE == 0);
+    // printf("%s ncols=%d, nrows=%d, WARP_SIZE=%d\n", __func__, ncols, nrows, WARP_SIZE);
+    if (ncols < 1024) {
+        const sycl::range<3> block_dims(1, 1, WARP_SIZE);
+        stream->submit([&](sycl::handler &cgh) {
+            sycl::local_accessor<float, 1> s_sum_acc_ct1(sycl::range<1>(32),
+                                                         cgh);
+
+            cgh.parallel_for(
+                sycl::nd_range<3>(sycl::range<3>(1, 1, nrows) * block_dims,
+                                  block_dims),
+                [=](sycl::nd_item<3> item_ct1)
+                    [[intel::reqd_sub_group_size(32)]] {
+                        rms_norm_f32(x, dst, ncols, eps, item_ct1,
+                                                s_sum_acc_ct1.get_pointer(), WARP_SIZE);
+                    });
+        });
+    } else {
+        const int work_group_size = get_work_group_size(stream->get_device());
+        const sycl::range<3> block_dims(1, 1, work_group_size);
+        /*
+        DPCT1049:19: The work-group size passed to the SYCL kernel may exceed
+        the limit. To get the device limit, query
+        info::device::max_work_group_size. Adjust the work-group size if needed.
+        */
+        stream->submit([&](sycl::handler &cgh) {
+            sycl::local_accessor<float, 1> s_sum_acc_ct1(sycl::range<1>(32),
+                                                         cgh);
+
+            cgh.parallel_for(
+                sycl::nd_range<3>(sycl::range<3>(1, 1, nrows) * block_dims,
+                                  block_dims),
+                [=](sycl::nd_item<3> item_ct1)
+                    [[intel::reqd_sub_group_size(32)]] {
+                        rms_norm_f32(x, dst, ncols, eps, item_ct1,
+                                           s_sum_acc_ct1.get_pointer(), work_group_size);
+                    });
+        });
+    }
+}
+
 static void quantize_row_q8_1_sycl(const float *x, void *vy, const int kx,
                                   const int ky, const int kx_padded,
                                   queue_ptr stream) {
    const int block_num_x = (kx_padded + SYCL_QUANTIZE_BLOCK_SIZE - 1) / SYCL_QUANTIZE_BLOCK_SIZE;
    const sycl::range<3> num_blocks(1, ky, block_num_x);
-    int constexpr QUANT_BLOCK_TILE = QK8_1 / WARP_SIZE;
-    static_assert(QK8_1 % WARP_SIZE == 0);
-    const sycl::range<3> block_size(1, 1, SYCL_QUANTIZE_BLOCK_SIZE / QUANT_BLOCK_TILE);
+    const sycl::range<3> block_size(1, 1, SYCL_DEQUANTIZE_BLOCK_SIZE);
    {
        dpct::has_capability_or_fail(stream->get_device(),
                                     {sycl::aspect::fp16});

        stream->parallel_for(
            sycl::nd_range<3>(num_blocks * block_size, block_size),
-            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                quantize_q8_1<QUANT_BLOCK_TILE>(x, vy, kx, kx_padded, item_ct1);
+            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
+                quantize_q8_1(x, vy, kx, kx_padded, item_ct1);
            });
    }
 }
@@ -1534,7 +1962,7 @@ static void ggml_mul_mat_p021_f16_f32_sycl(const void *vx, const float *y,

        stream->parallel_for(
            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
                mul_mat_p021_f16_f32(vx, y, dst, ncols_x, nrows_x, nchannels_x,
                                     nchannels_y, item_ct1);
            });
@@ -1554,7 +1982,7 @@ static void ggml_mul_mat_vec_nc_f16_f32_sycl(

        stream->parallel_for(
            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
                mul_mat_vec_nc_f16_f32(vx, y, dst, ncols_x, nrows_x,
                                       row_stride_x, channel_stride_x,
                                       nchannels_y / nchannels_x, item_ct1);
@@ -1813,13 +2241,117 @@ static void clamp_f32_sycl(const float *x, float *dst, const float min,
        });
 }

+template <typename T>
+static void rope_sycl(const T *x, T *dst, int ncols, int nrows,
+                      const int32_t *pos, float freq_scale, int p_delta_rows,
+                      float freq_base, float ext_factor, float attn_factor,
+                      rope_corr_dims corr_dims, queue_ptr stream) {
+    GGML_ASSERT(ncols % 2 == 0);
+    const sycl::range<3> block_dims(1, SYCL_ROPE_BLOCK_SIZE, 1);
+    const int num_blocks_x = (ncols + 2*SYCL_ROPE_BLOCK_SIZE - 1) / (2*SYCL_ROPE_BLOCK_SIZE);
+    const sycl::range<3> block_nums(1, num_blocks_x, nrows);
+    if (pos == nullptr) {
+        /*
+        DPCT1049:40: The work-group size passed to the SYCL kernel may exceed
+        the limit. To get the device limit, query
+        info::device::max_work_group_size. Adjust the work-group size if needed.
+        */
+        dpct::has_capability_or_fail(stream->get_device(),
+                                     {sycl::aspect::fp16});
+
+        stream->parallel_for(
+            sycl::nd_range<3>(block_nums * block_dims, block_dims),
+            [=](sycl::nd_item<3> item_ct1) {
+                rope<T, false>(x, dst, ncols, pos, freq_scale, p_delta_rows,
+                               freq_base, ext_factor, attn_factor, corr_dims,
+                               item_ct1);
+            });
+    } else {
+        /*
+        DPCT1049:41: The work-group size passed to the SYCL kernel may exceed
+        the limit. To get the device limit, query
+        info::device::max_work_group_size. Adjust the work-group size if needed.
+        */
+        dpct::has_capability_or_fail(stream->get_device(),
+                                     {sycl::aspect::fp16});
+
+        stream->parallel_for(
+            sycl::nd_range<3>(block_nums * block_dims, block_dims),
+            [=](sycl::nd_item<3> item_ct1) {
+                rope<T, true>(x, dst, ncols, pos, freq_scale, p_delta_rows,
+                              freq_base, ext_factor, attn_factor, corr_dims,
+                              item_ct1);
+            });
+    }
+}
+
+template <typename T>
+static void rope_neox_sycl(const T *x, T *dst, int ncols, int n_dims, int nrows,
+                           const int32_t *pos, float freq_scale,
+                           int p_delta_rows, float freq_base, float ext_factor,
+                           float attn_factor, rope_corr_dims corr_dims,
+                           const float * freq_factors, queue_ptr stream) {
+    GGML_ASSERT(ncols % 2 == 0);
+    const sycl::range<3> block_dims(1, SYCL_ROPE_BLOCK_SIZE, 1);
+    const int num_blocks_x = (ncols + 2*SYCL_ROPE_BLOCK_SIZE - 1) / (2*SYCL_ROPE_BLOCK_SIZE);
+    const sycl::range<3> block_nums(1, num_blocks_x, nrows);
+
+    const float theta_scale = powf(freq_base, -2.0f/n_dims);
+    const float inv_ndims = -1.0f / n_dims;
+
+    if (pos == nullptr) {
+        dpct::has_capability_or_fail(stream->get_device(),
+                                     {sycl::aspect::fp16});
+        if (freq_factors == nullptr) {
+            stream->parallel_for(
+                sycl::nd_range<3>(block_nums * block_dims, block_dims),
+                [=](sycl::nd_item<3> item_ct1) {
+                    rope_neox<T, false, false>(x, dst, ncols, n_dims, pos, freq_scale,
+                                        p_delta_rows, ext_factor, attn_factor,
+                                        corr_dims, theta_scale, inv_ndims, freq_factors,
+                                        item_ct1);
+                });
+        } else {
+            stream->parallel_for(
+                sycl::nd_range<3>(block_nums * block_dims, block_dims),
+                [=](sycl::nd_item<3> item_ct1) {
+                    rope_neox<T, false, true>(x, dst, ncols, n_dims, pos, freq_scale,
+                                        p_delta_rows, ext_factor, attn_factor,
+                                        corr_dims, theta_scale, inv_ndims, freq_factors,
+                                        item_ct1);
+                });
+        }
+    } else {
+        dpct::has_capability_or_fail(stream->get_device(),
+                                     {sycl::aspect::fp16});
+
+        if (freq_factors == nullptr) {
+            stream->parallel_for(
+                sycl::nd_range<3>(block_nums * block_dims, block_dims),
+                [=](sycl::nd_item<3> item_ct1) {
+                    rope_neox<T, true, false>(x, dst, ncols, n_dims, pos, freq_scale,
+                                       p_delta_rows, ext_factor, attn_factor,
+                                       corr_dims, theta_scale, inv_ndims, freq_factors, item_ct1);
+                });
+        } else {
+            stream->parallel_for(
+                sycl::nd_range<3>(block_nums * block_dims, block_dims),
+                [=](sycl::nd_item<3> item_ct1) {
+                    rope_neox<T, true, true>(x, dst, ncols, n_dims, pos, freq_scale,
+                                       p_delta_rows, ext_factor, attn_factor,
+                                       corr_dims, theta_scale, inv_ndims, freq_factors, item_ct1);
+                });
+        }
+    }
+}
+
 static void sum_rows_f32_sycl(const float *x, float *dst, const int ncols,
                              const int nrows, queue_ptr stream) {
    const sycl::range<3> block_dims(1, 1, WARP_SIZE);
    const sycl::range<3> block_nums(1, nrows, 1);
    stream->parallel_for(sycl::nd_range<3>(block_nums * block_dims, block_dims),
                         [=](sycl::nd_item<3> item_ct1)
-                             [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                             [[intel::reqd_sub_group_size(32)]] {
                                 k_sum_rows_f32(x, dst, ncols, item_ct1);
                             });
 }
@@ -1900,7 +2432,7 @@ static void soft_max_f32_submitter(const float * x, const float * mask, float *

        cgh.parallel_for(
            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
                soft_max_f32<vals_smem, ncols_template, block_size_template>(x, mask, dst, ncols_par,
                                                                             nrows_y, scale, max_bias, m0,
                                                                             m1, n_head_log2, item_ct1,
@@ -2080,6 +2612,12 @@ static inline int get_sycl_env(const char *env_name, int default_val) {
    return user_number;
 }

+static inline int get_work_group_size(const sycl::device& device) {
+    dpct::device_info prop;
+    dpct::get_device_info(prop, device);
+    return prop.get_max_work_group_size();
+}
+
 static void ggml_check_sycl() try {
    static bool initialized = false;

@@ -2638,6 +3176,45 @@ inline void ggml_sycl_op_sqr(ggml_backend_sycl_context & ctx, const ggml_tensor
    (void) src1_dd;
 }

+inline void ggml_sycl_op_norm(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
+                              ggml_tensor *dst, const float *src0_dd,
+                              const float *src1_dd, float *dst_dd,
+                              const queue_ptr &main_stream) {
+
+    GGML_ASSERT(src0->type == GGML_TYPE_F32);
+    GGML_ASSERT( dst->type == GGML_TYPE_F32);
+
+    const int64_t ne00 = src0->ne[0];
+    const int64_t nrows = ggml_nrows(src0);
+
+    float eps;
+    memcpy(&eps, dst->op_params, sizeof(float));
+
+    norm_f32_sycl(src0_dd, dst_dd, ne00, nrows, eps, main_stream);
+
+    (void) src1;
+    (void) dst;
+    (void) src1_dd;
+}
+
+inline void ggml_sycl_op_group_norm(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
+                                    const ggml_tensor *src1, ggml_tensor *dst,
+                                    const float *src0_dd, const float *src1_dd,
+                                    float *dst_dd,
+                                    const queue_ptr &main_stream) {
+
+    GGML_ASSERT(src0->type == GGML_TYPE_F32);
+    GGML_ASSERT( dst->type == GGML_TYPE_F32);
+
+    int num_groups = dst->op_params[0];
+    int group_size = src0->ne[0] * src0->ne[1] * ((src0->ne[2] + num_groups - 1) / num_groups);
+    group_norm_f32_sycl(src0_dd, dst_dd, num_groups, group_size, src0->ne[0] * src0->ne[1] * src0->ne[2], main_stream);
+
+    (void) src1;
+    (void) dst;
+    (void) src1_dd;
+}
+
 inline void ggml_sycl_op_concat(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
                                const ggml_tensor *src1, ggml_tensor *dst,
                                const float *src0_dd, const float *src1_dd,
@@ -2701,6 +3278,28 @@ inline void ggml_sycl_op_pad(ggml_backend_sycl_context & ctx, const ggml_tensor
    (void) src1_dd;
 }

+inline void ggml_sycl_op_rms_norm(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
+                                  const ggml_tensor *src1, ggml_tensor *dst,
+                                  const float *src0_dd, const float *src1_dd,
+                                  float *dst_dd,
+                                  const queue_ptr &main_stream) {
+
+    GGML_ASSERT(src0->type == GGML_TYPE_F32);
+    GGML_ASSERT( dst->type == GGML_TYPE_F32);
+
+    const int64_t ne00 = src0->ne[0];
+    const int64_t nrows = ggml_nrows(src0);
+
+    float eps;
+    memcpy(&eps, dst->op_params, sizeof(float));
+
+    rms_norm_f32_sycl(src0_dd, dst_dd, ne00, nrows, eps, main_stream);
+
+    (void) src1;
+    (void) dst;
+    (void) src1_dd;
+}
+
 static int64_t get_row_rounding(ggml_type type, const std::array<float, GGML_SYCL_MAX_DEVICES> & tensor_split) {
    int64_t min_compute_capability = INT_MAX;
    int64_t max_compute_capability = INT_MIN;
@@ -2862,6 +3461,97 @@ catch (sycl::exception const &exc) {
  std::exit(1);
 }

+inline void ggml_sycl_op_rope(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
+                              ggml_tensor *dst, const float *src0_dd,
+                              const float *src1_dd, float *dst_dd,
+                              const queue_ptr &main_stream) {
+    const ggml_tensor * src2 = dst->src[2];
+
+    GGML_ASSERT(src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_F16);
+    GGML_ASSERT( dst->type == GGML_TYPE_F32 ||  dst->type == GGML_TYPE_F16);
+    GGML_ASSERT(src0->type == dst->type);
+
+    const int64_t ne00 = src0->ne[0];
+    const int64_t ne01 = src0->ne[1];
+    const int64_t ne2 = dst->ne[2];
+    const int64_t nrows = ggml_nrows(src0);
+
+    //const int n_past      = ((int32_t *) dst->op_params)[0];
+    const int n_dims      = ((int32_t *) dst->op_params)[1];
+    const int mode        = ((int32_t *) dst->op_params)[2];
+    //const int n_ctx       = ((int32_t *) dst->op_params)[3];
+    const int n_ctx_orig  = ((int32_t *) dst->op_params)[4];
+
+    // RoPE alteration for extended context
+    float freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow;
+    memcpy(&freq_base,   (int32_t *) dst->op_params +  5, sizeof(float));
+    memcpy(&freq_scale,  (int32_t *) dst->op_params +  6, sizeof(float));
+    memcpy(&ext_factor,  (int32_t *) dst->op_params +  7, sizeof(float));
+    memcpy(&attn_factor, (int32_t *) dst->op_params +  8, sizeof(float));
+    memcpy(&beta_fast,   (int32_t *) dst->op_params +  9, sizeof(float));
+    memcpy(&beta_slow,   (int32_t *) dst->op_params + 10, sizeof(float));
+
+    const float * freq_factors = nullptr;
+    const int32_t * pos = nullptr;
+    if ((mode & 1) == 0) {
+        GGML_ASSERT(src1->type == GGML_TYPE_I32);
+        GGML_ASSERT(src1->ne[0] == ne2);
+        pos = (const int32_t *) src1_dd;
+    }
+
+    const bool is_neox = mode & 2;
+
+#pragma message("TODO: update rope NORM mode to match NEOX mode")
+#pragma message("      https://github.com/ggerganov/llama.cpp/pull/7634")
+
+    if (is_neox) {
+        pos = (const int32_t *) src1_dd;
+
+        if (src2 != nullptr) {
+            freq_factors = (const float *) src2->data;
+        }
+    } else {
+        GGML_ASSERT(src2 == nullptr && "TODO: freq_factors not implemented for !is_neox");
+    }
+
+    rope_corr_dims corr_dims;
+    ggml_rope_yarn_corr_dims(n_dims, n_ctx_orig, freq_base, beta_fast, beta_slow, corr_dims.v);
+
+    // compute
+    if (is_neox) {
+        if (src0->type == GGML_TYPE_F32) {
+            rope_neox_sycl(
+                (const float *)src0_dd, (float *)dst_dd, ne00, n_dims, nrows, pos, freq_scale, ne01, freq_base, ext_factor,
+                attn_factor, corr_dims, freq_factors, main_stream
+            );
+        } else if (src0->type == GGML_TYPE_F16) {
+            rope_neox_sycl((const sycl::half *)src0_dd, (sycl::half *)dst_dd,
+                           ne00, n_dims, nrows, pos, freq_scale, ne01,
+                           freq_base, ext_factor, attn_factor, corr_dims,
+                           freq_factors, main_stream);
+        } else {
+            GGML_ASSERT(false);
+        }
+    } else {
+        if (src0->type == GGML_TYPE_F32) {
+            rope_sycl(
+                (const float *)src0_dd, (float *)dst_dd, ne00, nrows, pos, freq_scale, ne01, freq_base, ext_factor,
+                attn_factor, corr_dims, main_stream
+            );
+        } else if (src0->type == GGML_TYPE_F16) {
+            rope_sycl((const sycl::half *)src0_dd, (sycl::half *)dst_dd, ne00,
+                      nrows, pos, freq_scale, ne01, freq_base, ext_factor,
+                      attn_factor, corr_dims, main_stream);
+        } else {
+            GGML_ASSERT(false);
+        }
+    }
+
+    (void) src1;
+    (void) dst;
+    (void) src1_dd;
+}
+
 static void ggml_sycl_op_pool2d(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
                                const ggml_tensor *src1, ggml_tensor *dst,
                                const float *src0_dd, const float *src1_dd,
@@ -3886,6 +4576,7 @@ bool ggml_sycl_supports_dmmv(enum ggml_type type) {

 static void ggml_sycl_mul_mat(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
    const bool split = ggml_backend_buffer_is_sycl_split(src0->buffer);
+
    int64_t min_compute_capability = INT_MAX;

    if (split) {
@@ -5550,9 +6241,7 @@ GGML_CALL static bool ggml_backend_sycl_supports_op(ggml_backend_t backend, cons
        case GGML_OP_CONT:
        case GGML_OP_DIAG_MASK_INF:
        case GGML_OP_SOFT_MAX:
-            return true;
        case GGML_OP_ROPE:
-            return ggml_is_contiguous(op->src[0]);
        case GGML_OP_IM2COL:
        case GGML_OP_POOL_2D:
        case GGML_OP_SUM_ROWS:
@@ -19,7 +19,5 @@
 #include "dmmv.hpp"
 #include "mmq.hpp"
 #include "mmvq.hpp"
-#include "rope.hpp"
-#include "norm.hpp"

 #endif // GGML_SYCL_BACKEND_HPP
@@ -295,66 +295,5 @@ struct ggml_backend_sycl_context {
    }
 };

-// common host functions
-
-static inline int get_work_group_size(const sycl::device& device) {
-    dpct::device_info prop;
-    dpct::get_device_info(prop, device);
-    return prop.get_max_work_group_size();
-}
-
-
-// common device functions
-
-static __dpct_inline__ float warp_reduce_sum(float x,
-    const sycl::nd_item<3>& item_ct1) {
-#pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
-        /*
-        DPCT1096:98: The right-most dimension of the work-group used in the SYCL
-        kernel that calls this function may be less than "32". The function
-        "dpct::permute_sub_group_by_xor" may return an unexpected result on the
-        CPU device. Modify the size of the work-group to ensure that the value
-        of the right-most dimension is a multiple of "32".
-        */
-        x += dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), x, mask);
-    }
-    return x;
-}
-
-static __dpct_inline__ sycl::float2
-warp_reduce_sum(sycl::float2 a, const sycl::nd_item<3>& item_ct1) {
-#pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
-        a.x() += dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), a.x(),
-            mask);
-        a.y() += dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), a.y(),
-            mask);
-    }
-    return a;
-}
-
-static __dpct_inline__ float warp_reduce_max(float x,
-    const sycl::nd_item<3>& item_ct1) {
-#pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
-        /*
-        DPCT1096:97: The right-most dimension of the work-group used in the SYCL
-        kernel that calls this function may be less than "32". The function
-        "dpct::permute_sub_group_by_xor" may return an unexpected result on the
-        CPU device. Modify the size of the work-group to ensure that the value
-        of the right-most dimension is a multiple of "32".
-        */
-        x = sycl::fmax(x, dpct::permute_sub_group_by_xor(
-            item_ct1.get_sub_group(), x, mask));
-    }
-    return x;
-}
-
-// Helper for vec loading aligned data
-template <typename Tp, int n>
-inline sycl::vec<Tp, n> vec_aligned_load(const Tp* aligned_ptr) {
-    return *reinterpret_cast<const sycl::vec<Tp, n>*>(aligned_ptr);
-}

 #endif // GGML_SYCL_COMMON_HPP
@@ -152,15 +152,12 @@ static void dequantize_row_q4_K_sycl(const void *vx, dst_t *y, const int k,
        dpct::has_capability_or_fail(stream->get_device(),
                                     {sycl::aspect::fp16});

-        stream->submit([&](sycl::handler &cgh) {
-            sycl::local_accessor<uint8_t, 1> scale_local_acc(sycl::range<1>(12), cgh);
-            cgh.parallel_for(sycl::nd_range<3>(sycl::range<3>(1, 1, nb) *
+        stream->parallel_for(sycl::nd_range<3>(sycl::range<3>(1, 1, nb) *
                                                   sycl::range<3>(1, 1, 32),
                                               sycl::range<3>(1, 1, 32)),
                             [=](sycl::nd_item<3> item_ct1) {
-                                 dequantize_block_q4_K(vx, y, scale_local_acc.get_pointer(), item_ct1);
+                                 dequantize_block_q4_K(vx, y, item_ct1);
                             });
-        });
    }
 }

@@ -293,8 +293,7 @@ static void dequantize_block_q3_K(const void * __restrict__ vx, dst_t * __restri
 #if QK_K == 256
 static inline void get_scale_min_k4(int j, const uint8_t * q, uint8_t & d, uint8_t & m) {
    if (j < 4) {
-        d = q[j] & 63;
-        m = q[j + 4] & 63;
+        d = q[j] & 63; m = q[j + 4] & 63;
    } else {
        d = (q[j+4] & 0xF) | ((q[j-4] >> 6) << 4);
        m = (q[j+4] >>  4) | ((q[j-0] >> 6) << 4);
@@ -304,7 +303,7 @@ static inline void get_scale_min_k4(int j, const uint8_t * q, uint8_t & d, uint8

 template<typename dst_t>
 static void dequantize_block_q4_K(const void * __restrict__ vx, dst_t * __restrict__ yy,
-                                  uint8_t* scales_local, const sycl::nd_item<3> &item_ct1) {
+                                  const sycl::nd_item<3> &item_ct1) {
    const block_q4_K * x = (const block_q4_K *) vx;

    const int i = item_ct1.get_group(2);
@@ -319,26 +318,19 @@ static void dequantize_block_q4_K(const void * __restrict__ vx, dst_t * __restri

    dst_t * y = yy + i*QK_K + 64*il + n*ir;

-    const sycl::half2 dm = x[i].dm;
-    const float dall = dm[0];
-    const float dmin = dm[1];
+    const float dall = x[i].dm[0];
+    const float dmin = x[i].dm[1];

-    if (tid < 12)
-        scales_local[tid] = x[i].scales[tid];
-    item_ct1.barrier(sycl::access::fence_space::local_space);
+    const uint8_t * q = x[i].qs + 32*il + n*ir;

    uint8_t sc, m;
-    get_scale_min_k4(is + 0, scales_local, sc, m);
-    const float d1 = dall * sc;
-    const float m1 = dmin * m;
-    get_scale_min_k4(is + 1, scales_local, sc, m);
-    const float d2 = dall * sc;
-    const float m2 = dmin * m;
-
-    sycl::vec<uint8_t, n> q_vec = vec_aligned_load<uint8_t, n>(x[i].qs + 32*il + n*ir);
+    get_scale_min_k4(is + 0, x[i].scales, sc, m);
+    const float d1 = dall * sc; const float m1 = dmin * m;
+    get_scale_min_k4(is + 1, x[i].scales, sc, m);
+    const float d2 = dall * sc; const float m2 = dmin * m;
    for (int l = 0; l < n; ++l) {
-        y[l + 0] = d1 * (q_vec[l] & 0xF) - m1;
-        y[l +32] = d2 * (q_vec[l] >>  4) - m2;
+        y[l + 0] = d1 * (q[l] & 0xF) - m1;
+        y[l +32] = d2 * (q[l] >>  4) - m2;
    }
 #else
    const int tid = item_ct1.get_local_id(2);
@@ -76,7 +76,7 @@ static void dequantize_mul_mat_vec(const void * __restrict__ vx, const dfloat *

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -104,7 +104,7 @@ static void convert_mul_mat_vec_f16_sycl(const void *vx, const dfloat *y,

        stream->parallel_for(
            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
                dequantize_mul_mat_vec<1, 1, convert_f16>(vx, y, dst, ncols,
                                                          nrows, item_ct1);
            });
@@ -227,7 +227,7 @@ static void dequantize_mul_mat_vec_q2_k(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -346,7 +346,7 @@ static void dequantize_mul_mat_vec_q3_k(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -499,7 +499,7 @@ static void dequantize_mul_mat_vec_q4_k(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -633,7 +633,7 @@ static void dequantize_mul_mat_vec_q5_k(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -748,7 +748,7 @@ static void dequantize_mul_mat_vec_q6_k(const void * __restrict__ vx, const floa

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -774,7 +774,7 @@ static void dequantize_mul_mat_vec_q4_0_sycl(const void *vx, const dfloat *y,

        stream->parallel_for(
            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
                dequantize_mul_mat_vec<QK4_0, QR4_0, dequantize_q4_0>(
                    vx, y, dst, ncols, nrows, item_ct1);
            });
@@ -795,7 +795,7 @@ static void dequantize_mul_mat_vec_q4_1_sycl(const void *vx, const dfloat *y,

        stream->parallel_for(
            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
                dequantize_mul_mat_vec<QK4_1, QR4_1, dequantize_q4_1>(
                    vx, y, dst, ncols, nrows, item_ct1);
            });
@@ -816,7 +816,7 @@ static void dequantize_mul_mat_vec_q5_0_sycl(const void *vx, const dfloat *y,

        stream->parallel_for(
            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
                dequantize_mul_mat_vec<QK5_0, QR5_0, dequantize_q5_0>(
                    vx, y, dst, ncols, nrows, item_ct1);
            });
@@ -837,7 +837,7 @@ static void dequantize_mul_mat_vec_q5_1_sycl(const void *vx, const dfloat *y,

        stream->parallel_for(
            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
                dequantize_mul_mat_vec<QK5_1, QR5_1, dequantize_q5_1>(
                    vx, y, dst, ncols, nrows, item_ct1);
            });
@@ -858,7 +858,7 @@ static void dequantize_mul_mat_vec_q8_0_sycl(const void *vx, const dfloat *y,

        stream->parallel_for(
            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+            [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
                dequantize_mul_mat_vec<QK8_0, QR8_0, dequantize_q8_0>(
                    vx, y, dst, ncols, nrows, item_ct1);
            });
@@ -873,10 +873,10 @@ static void dequantize_mul_mat_vec_q2_K_sycl(const void *vx, const float *y,
    const int ny = 2; // very slightly faster than 1 even when K_QUANTS_PER_ITERATION = 2
    const int block_num_y = (nrows + ny - 1) / ny;
    const sycl::range<3> block_nums(1, 1, block_num_y);
-    const sycl::range<3> block_dims(1, ny, WARP_SIZE);
+    const sycl::range<3> block_dims(1, ny, 32);
    stream->parallel_for(
        sycl::nd_range<3>(block_nums * block_dims, block_dims),
-        [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+        [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
            dequantize_mul_mat_vec_q2_k(vx, y, dst, ncols, nrows, item_ct1);
        });
 }
@@ -889,10 +889,10 @@ static void dequantize_mul_mat_vec_q3_K_sycl(const void *vx, const float *y,
    const int ny = 2 / K_QUANTS_PER_ITERATION;
    const int block_num_y = (nrows + ny - 1) / ny;
    const sycl::range<3> block_nums(1, 1, block_num_y);
-    const sycl::range<3> block_dims(1, ny, WARP_SIZE);
+    const sycl::range<3> block_dims(1, ny, 32);
    stream->parallel_for(
        sycl::nd_range<3>(block_nums * block_dims, block_dims),
-        [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+        [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
            dequantize_mul_mat_vec_q3_k(vx, y, dst, ncols, nrows, item_ct1);
        });
 }
@@ -905,10 +905,10 @@ static void dequantize_mul_mat_vec_q4_K_sycl(const void *vx, const float *y,
    const int ny = 2 / K_QUANTS_PER_ITERATION;
    const int block_num_y = (nrows + ny - 1) / ny;
    const sycl::range<3> block_nums(1, 1, block_num_y);
-    const sycl::range<3> block_dims(1, ny, WARP_SIZE);
+    const sycl::range<3> block_dims(1, ny, 32);
    stream->parallel_for(
        sycl::nd_range<3>(block_nums * block_dims, block_dims),
-        [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+        [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
            dequantize_mul_mat_vec_q4_k(vx, y, dst, ncols, nrows, item_ct1);
        });
 }
@@ -918,10 +918,10 @@ static void dequantize_mul_mat_vec_q5_K_sycl(const void *vx, const float *y,
                                             const int nrows,
                                             dpct::queue_ptr stream) {
    GGML_ASSERT(ncols % QK_K == 0);
-    const sycl::range<3> block_dims(1, 1, WARP_SIZE);
+    const sycl::range<3> block_dims(1, 1, 32);
    stream->parallel_for(
        sycl::nd_range<3>(sycl::range<3>(1, 1, nrows) * block_dims, block_dims),
-        [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+        [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
            dequantize_mul_mat_vec_q5_k(vx, y, dst, ncols, item_ct1);
        });
 }
@@ -934,10 +934,10 @@ static void dequantize_mul_mat_vec_q6_K_sycl(const void *vx, const float *y,
    const int ny = 2 / K_QUANTS_PER_ITERATION;
    const int block_num_y = (nrows + ny - 1) / ny;
    const sycl::range<3> block_nums(1, 1, block_num_y);
-    const sycl::range<3> block_dims(1, ny, WARP_SIZE);
+    const sycl::range<3> block_dims(1, ny, 32);
    stream->parallel_for(
        sycl::nd_range<3>(block_nums * block_dims, block_dims),
-        [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+        [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
            dequantize_mul_mat_vec_q6_k(vx, y, dst, ncols, nrows, item_ct1);
        });
 }
@@ -37,7 +37,7 @@ static void mul_mat_vec_q(const void * __restrict__ vx, const void * __restrict_

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -85,7 +85,7 @@ static void mul_mat_vec_q_iq2_xxs_q8_1(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -133,7 +133,7 @@ static void mul_mat_vec_q_iq2_xs_q8_1(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -181,7 +181,7 @@ static void mul_mat_vec_q_iq2_s_q8_1(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -229,7 +229,7 @@ static void mul_mat_vec_q_iq3_xxs_q8_1(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -277,7 +277,7 @@ static void mul_mat_vec_q_iq3_s_q8_1(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -325,7 +325,7 @@ static void mul_mat_vec_q_iq1_s_q8_1(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -373,7 +373,7 @@ static void mul_mat_vec_q_iq1_m_q8_1(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -421,7 +421,7 @@ static void mul_mat_vec_q_iq4_nl_q8_1(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -470,7 +470,7 @@ static void mul_mat_vec_q_iq4_xs_q8_1(const void *__restrict__ vx,

    // sum up partial sums and write back result
 #pragma unroll
-    for (int mask = WARP_SIZE / 2; mask > 0; mask >>= 1) {
+    for (int mask = 16; mask > 0; mask >>= 1) {
        tmp +=
            dpct::permute_sub_group_by_xor(item_ct1.get_sub_group(), tmp, mask);
    }
@@ -495,7 +495,7 @@ static void mul_mat_vec_q4_0_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q<QK4_0, QI4_0, block_q4_0,
                                      VDR_Q4_0_Q8_1_MMVQ, vec_dot_q4_0_q8_1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
@@ -519,7 +519,7 @@ static void mul_mat_vec_q4_1_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q<QK4_0, QI4_1, block_q4_1,
                                      VDR_Q4_1_Q8_1_MMVQ, vec_dot_q4_1_q8_1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
@@ -543,7 +543,7 @@ static void mul_mat_vec_q5_0_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q<QK5_0, QI5_0, block_q5_0,
                                      VDR_Q5_0_Q8_1_MMVQ, vec_dot_q5_0_q8_1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
@@ -567,7 +567,7 @@ static void mul_mat_vec_q5_1_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q<QK5_1, QI5_1, block_q5_1,
                                      VDR_Q5_1_Q8_1_MMVQ, vec_dot_q5_1_q8_1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
@@ -591,7 +591,7 @@ static void mul_mat_vec_q8_0_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q<QK8_0, QI8_0, block_q8_0,
                                      VDR_Q8_0_Q8_1_MMVQ, vec_dot_q8_0_q8_1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
@@ -615,7 +615,7 @@ static void mul_mat_vec_q2_K_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q<QK_K, QI2_K, block_q2_K,
                                      VDR_Q2_K_Q8_1_MMVQ, vec_dot_q2_K_q8_1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
@@ -639,7 +639,7 @@ static void mul_mat_vec_q3_K_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q<QK_K, QI3_K, block_q3_K,
                                      VDR_Q3_K_Q8_1_MMVQ, vec_dot_q3_K_q8_1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
@@ -663,7 +663,7 @@ static void mul_mat_vec_q4_K_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q<QK_K, QI4_K, block_q4_K,
                                      VDR_Q4_K_Q8_1_MMVQ, vec_dot_q4_K_q8_1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
@@ -687,7 +687,7 @@ static void mul_mat_vec_q5_K_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q<QK_K, QI5_K, block_q5_K,
                                      VDR_Q5_K_Q8_1_MMVQ, vec_dot_q5_K_q8_1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
@@ -711,7 +711,7 @@ static void mul_mat_vec_q6_K_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q<QK_K, QI6_K, block_q6_K,
                                      VDR_Q6_K_Q8_1_MMVQ, vec_dot_q6_K_q8_1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
@@ -734,8 +734,8 @@ static void mul_mat_vec_iq2_xxs_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                        mul_mat_vec_q_iq2_xxs_q8_1<QK_K, QI2_XXS/2, block_iq2_xxs, 1>(
+                    [[intel::reqd_sub_group_size(32)]] {
+                        mul_mat_vec_q_iq2_xxs_q8_1<QK_K, QI2_XXS, block_iq2_xxs, 1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
                    });
        });
@@ -759,8 +759,8 @@ static void mul_mat_vec_iq2_xs_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                        mul_mat_vec_q_iq2_xs_q8_1<QK_K, QI2_XS/2, block_iq2_xs, 1>(
+                    [[intel::reqd_sub_group_size(32)]] {
+                        mul_mat_vec_q_iq2_xs_q8_1<QK_K, QI2_XS, block_iq2_xs, 1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
                    });
        });
@@ -784,8 +784,8 @@ static void mul_mat_vec_iq2_s_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                        mul_mat_vec_q_iq2_s_q8_1<QK_K, QI2_S/2, block_iq2_s, 1>(
+                    [[intel::reqd_sub_group_size(32)]] {
+                        mul_mat_vec_q_iq2_s_q8_1<QK_K, QI2_S, block_iq2_s, 1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
                    });
        });
@@ -809,8 +809,8 @@ static void mul_mat_vec_iq3_xxs_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                        mul_mat_vec_q_iq3_xxs_q8_1<QK_K, QI3_XXS/2, block_iq3_xxs, 1>(
+                    [[intel::reqd_sub_group_size(32)]] {
+                        mul_mat_vec_q_iq3_xxs_q8_1<QK_K, QI3_XXS, block_iq3_xxs, 1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
                    });
        });
@@ -833,8 +833,8 @@ static void mul_mat_vec_iq3_s_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                        mul_mat_vec_q_iq3_s_q8_1<QK_K, QI3_S/2, block_iq3_s, 1>(
+                    [[intel::reqd_sub_group_size(32)]] {
+                        mul_mat_vec_q_iq3_s_q8_1<QK_K, QI3_XS, block_iq3_s, 1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
                    });
        });
@@ -858,7 +858,7 @@ static void mul_mat_vec_iq1_s_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q_iq1_s_q8_1<QK_K, QI1_S, block_iq1_s, 1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
                    });
@@ -879,7 +879,7 @@ static void mul_mat_vec_iq1_m_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q_iq1_m_q8_1<QK_K, QI1_S, block_iq1_m, 1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
                    });
@@ -901,7 +901,7 @@ static void mul_mat_vec_iq4_nl_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
+                    [[intel::reqd_sub_group_size(32)]] {
                        mul_mat_vec_q_iq4_nl_q8_1<QK4_NL, QI4_NL, block_iq4_nl, 1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
                    });
@@ -923,8 +923,8 @@ static void mul_mat_vec_iq4_xs_q8_1_sycl(const void *vx, const void *vy,
            cgh.parallel_for(
                sycl::nd_range<3>(block_nums * block_dims, block_dims),
                [=](sycl::nd_item<3> item_ct1)
-                    [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                        mul_mat_vec_q_iq4_xs_q8_1<QK_K, QI4_XS/4, block_iq4_xs, 1>(
+                    [[intel::reqd_sub_group_size(32)]] {
+                        mul_mat_vec_q_iq4_xs_q8_1<QK_K, QI4_XS, block_iq4_xs, 1>(
                            vx, vy, dst, ncols, nrows, item_ct1);
                    });
        });
@@ -936,7 +936,7 @@ void ggml_sycl_op_mul_mat_vec_q(
    const ggml_tensor *src0, const ggml_tensor *src1, ggml_tensor *dst,
    const char *src0_dd_i, const float *src1_ddf_i, const char *src1_ddq_i,
    float *dst_dd_i, const int64_t row_low, const int64_t row_high,
-    const int64_t src1_ncols, const int64_t src1_padded_col_size,
+    const int64_t src1_ncols, const int64_t src1_padded_row_size,
    const dpct::queue_ptr &stream) {

    const int64_t ne10 = src1->ne[0];
@@ -948,80 +948,77 @@ void ggml_sycl_op_mul_mat_vec_q(
    int id;
    SYCL_CHECK(
        CHECK_TRY_ERROR(id = get_current_device_id()));
-    const size_t q8_1_ts = sizeof(block_q8_1);
-    const size_t q8_1_bs = QK8_1;
+
    // the main device has a larger memory buffer to hold the results from all GPUs
    // nrows_dst == nrows of the matrix that the kernel writes into
    const int64_t nrows_dst = id == ctx.device ? ne00 : row_diff;
-    for (int i = 0; i < src1_ncols; i++)
-    {
-        const size_t src1_ddq_i_offset = i * src1_padded_col_size * q8_1_ts / q8_1_bs;
-        const char* src1_ddq_i_bs = src1_ddq_i + src1_ddq_i_offset;
-        float* dst_dd_i_bs = dst_dd_i + i * dst->ne[0];
-        switch (src0->type) {
+
+    switch (src0->type) {
        case GGML_TYPE_Q4_0:
-            mul_mat_vec_q4_0_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_q4_0_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_Q4_1:
-            mul_mat_vec_q4_1_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_q4_1_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_Q5_0:
-            mul_mat_vec_q5_0_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_q5_0_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_Q5_1:
-            mul_mat_vec_q5_1_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_q5_1_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_Q8_0:
-            mul_mat_vec_q8_0_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_q8_0_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_Q2_K:
-            mul_mat_vec_q2_K_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_q2_K_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_Q3_K:
-            mul_mat_vec_q3_K_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_q3_K_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_Q4_K:
-            mul_mat_vec_q4_K_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_q4_K_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_Q5_K:
-            mul_mat_vec_q5_K_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_q5_K_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_Q6_K:
-            mul_mat_vec_q6_K_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_q6_K_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_IQ1_S:
-            mul_mat_vec_iq1_s_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_iq1_s_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_IQ1_M:
-            mul_mat_vec_iq1_m_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_iq1_m_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_IQ2_XXS:
-            mul_mat_vec_iq2_xxs_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_iq2_xxs_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_IQ2_XS:
-            mul_mat_vec_iq2_xs_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_iq2_xs_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_IQ2_S:
-            mul_mat_vec_iq2_s_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_iq2_s_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_IQ3_XXS:
-            mul_mat_vec_iq3_xxs_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_iq3_xxs_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_IQ3_S:
-            mul_mat_vec_iq3_s_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_iq3_s_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_IQ4_NL:
-            mul_mat_vec_iq4_nl_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_iq4_nl_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        case GGML_TYPE_IQ4_XS:
-            mul_mat_vec_iq4_xs_q8_1_sycl(src0_dd_i, src1_ddq_i_bs, dst_dd_i_bs, ne00, row_diff, stream);
+            mul_mat_vec_iq4_xs_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, stream);
            break;
        default:
            GGML_ASSERT(false);
            break;
-        }
    }
+
    (void) src1;
    (void) dst;
    (void) src1_ddf_i;
+    (void) src1_ncols;
+    (void) src1_padded_row_size;
 }
@@ -1,370 +0,0 @@
-#include "norm.hpp"
-
-static void norm_f32(const float* x, float* dst, const int ncols, const float eps,
-    const sycl::nd_item<3>& item_ct1, sycl::float2* s_sum, int block_size) {
-    const int row = item_ct1.get_group(2) * item_ct1.get_local_range(1) +
-        item_ct1.get_local_id(1);
-    const int tid = item_ct1.get_local_id(2);
-
-    const int nthreads = item_ct1.get_local_range(2);
-    const int nwarps = nthreads / WARP_SIZE;
-    assert(nwarps % WARP_SIZE == 0);
-    sycl::float2 mean_var = sycl::float2(0.f, 0.f);
-
-    for (int col = tid; col < ncols; col += block_size) {
-        const float xi = x[row * ncols + col];
-        mean_var.x() += xi;
-        mean_var.y() += xi * xi;
-    }
-
-    // sum up partial sums
-    mean_var = warp_reduce_sum(mean_var, item_ct1);
-    if (block_size > WARP_SIZE) {
-
-        int warp_id = item_ct1.get_local_id(2) / WARP_SIZE;
-        int lane_id = item_ct1.get_local_id(2) % WARP_SIZE;
-        if (lane_id == 0) {
-            s_sum[warp_id] = mean_var;
-        }
-        /*
-        DPCT1118:0: SYCL group functions and algorithms must be encountered in
-        converged control flow. You may need to adjust the code.
-        */
-        item_ct1.barrier(sycl::access::fence_space::local_space);
-        mean_var = 0.f;
-        int nreduce = nwarps / WARP_SIZE;
-        for (size_t i = 0; i < nreduce; i += 1)
-        {
-            mean_var += s_sum[lane_id + i * WARP_SIZE];
-        }
-        mean_var = warp_reduce_sum(mean_var, item_ct1);
-    }
-
-    const float mean = mean_var.x() / ncols;
-    const float var = mean_var.y() / ncols - mean * mean;
-    const float inv_std = sycl::rsqrt(var + eps);
-
-    for (int col = tid; col < ncols; col += block_size) {
-        dst[row * ncols + col] = (x[row * ncols + col] - mean) * inv_std;
-    }
-}
-
-static void group_norm_f32(const float* x, float* dst, const int group_size, const int ne_elements, const float eps,
-    const sycl::nd_item<3>& item_ct1, float* s_sum, int block_size) {
-    int start = item_ct1.get_group(2) * group_size;
-    int end = start + group_size;
-    const int nthreads = item_ct1.get_local_range(2);
-    const int nwarps = nthreads / WARP_SIZE;
-    assert(nwarps % WARP_SIZE == 0);
-    start += item_ct1.get_local_id(2);
-
-    if (end >= ne_elements) {
-        end = ne_elements;
-    }
-
-    float tmp = 0.0f; // partial sum for thread in warp
-
-    for (int j = start; j < end; j += block_size) {
-        tmp += x[j];
-    }
-
-    tmp = warp_reduce_sum(tmp, item_ct1);
-    if (block_size > WARP_SIZE) {
-
-        int warp_id = item_ct1.get_local_id(2) / WARP_SIZE;
-        int lane_id = item_ct1.get_local_id(2) % WARP_SIZE;
-        if (lane_id == 0) {
-            s_sum[warp_id] = tmp;
-        }
-        /*
-        DPCT1118:1: SYCL group functions and algorithms must be encountered in
-        converged control flow. You may need to adjust the code.
-        */
-        /*
-        DPCT1065:54: Consider replacing sycl::nd_item::barrier() with
-        sycl::nd_item::barrier(sycl::access::fence_space::local_space) for
-        better performance if there is no access to global memory.
-        */
-        item_ct1.barrier();
-        tmp = 0.f;
-        int nreduce = nwarps / WARP_SIZE;
-        for (size_t i = 0; i < nreduce; i += 1)
-        {
-            tmp += s_sum[lane_id + i * WARP_SIZE];
-        }
-        tmp = warp_reduce_sum(tmp, item_ct1);
-    }
-
-    float mean = tmp / group_size;
-    tmp = 0.0f;
-
-    for (int j = start; j < end; j += block_size) {
-        float xi = x[j] - mean;
-        dst[j] = xi;
-        tmp += xi * xi;
-    }
-
-    tmp = warp_reduce_sum(tmp, item_ct1);
-    if (block_size > WARP_SIZE) {
-
-        int warp_id = item_ct1.get_local_id(2) / WARP_SIZE;
-        int lane_id = item_ct1.get_local_id(2) % WARP_SIZE;
-        if (lane_id == 0) {
-            s_sum[warp_id] = tmp;
-        }
-        /*
-        DPCT1118:2: SYCL group functions and algorithms must be encountered in
-        converged control flow. You may need to adjust the code.
-        */
-        /*
-        DPCT1065:55: Consider replacing sycl::nd_item::barrier() with
-        sycl::nd_item::barrier(sycl::access::fence_space::local_space) for
-        better performance if there is no access to global memory.
-        */
-        item_ct1.barrier();
-        tmp = s_sum[lane_id];
-        tmp = warp_reduce_sum(tmp, item_ct1);
-    }
-
-    float variance = tmp / group_size;
-    float scale = sycl::rsqrt(variance + eps);
-    for (int j = start; j < end; j += block_size) {
-        dst[j] *= scale;
-    }
-}
-
-static void rms_norm_f32(const float* x, float* dst, const int ncols, const float eps,
-    const sycl::nd_item<3>& item_ct1, float* s_sum, int block_size) {
-    const int row = item_ct1.get_group(2) * item_ct1.get_local_range(1) +
-        item_ct1.get_local_id(1);
-    const int tid = item_ct1.get_local_id(2);
-    const int nthreads = item_ct1.get_local_range(2);
-    const int nwarps = nthreads / WARP_SIZE;
-    assert(nwarps % WARP_SIZE == 0);
-    float tmp = 0.0f; // partial sum for thread in warp
-
-    for (int col = tid; col < ncols; col += block_size) {
-        const float xi = x[row * ncols + col];
-        tmp += xi * xi;
-    }
-
-    // sum up partial sums
-    tmp = warp_reduce_sum(tmp, item_ct1);
-    if (block_size > WARP_SIZE) {
-
-        int warp_id = item_ct1.get_local_id(2) / WARP_SIZE;
-        int lane_id = item_ct1.get_local_id(2) % WARP_SIZE;
-        if (lane_id == 0) {
-            s_sum[warp_id] = tmp;
-        }
-        /*
-        DPCT1118:3: SYCL group functions and algorithms must be encountered in
-        converged control flow. You may need to adjust the code.
-        */
-        item_ct1.barrier(sycl::access::fence_space::local_space);
-        int nreduce = nwarps / WARP_SIZE;
-        tmp = 0.f;
-        for (size_t i = 0; i < nreduce; i += 1)
-        {
-            tmp += s_sum[lane_id + i * WARP_SIZE];
-        }
-        tmp = warp_reduce_sum(tmp, item_ct1);
-    }
-
-    const float mean = tmp / ncols;
-    const float scale = sycl::rsqrt(mean + eps);
-
-    for (int col = tid; col < ncols; col += block_size) {
-        dst[row * ncols + col] = scale * x[row * ncols + col];
-    }
-}
-
-static void norm_f32_sycl(const float* x, float* dst, const int ncols,
-    const int nrows, const float eps,
-    queue_ptr stream) {
-    GGML_ASSERT(ncols % WARP_SIZE == 0);
-    if (ncols < 1024) {
-        const sycl::range<3> block_dims(1, 1, WARP_SIZE);
-        stream->submit([&](sycl::handler& cgh) {
-            cgh.parallel_for(
-                sycl::nd_range<3>(sycl::range<3>(1, 1, nrows) * block_dims,
-                    block_dims),
-                [=](sycl::nd_item<3> item_ct1)
-                [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                    norm_f32(x, dst, ncols, eps, item_ct1,
-                        nullptr, WARP_SIZE);
-                });
-            });
-    }
-    else {
-        const int work_group_size = get_work_group_size(stream->get_device());
-        const sycl::range<3> block_dims(1, 1, work_group_size);
-        /*
-        DPCT1049:17: The work-group size passed to the SYCL kernel may exceed
-        the limit. To get the device limit, query
-        info::device::max_work_group_size. Adjust the work-group size if needed.
-        */
-        stream->submit([&](sycl::handler& cgh) {
-            sycl::local_accessor<sycl::float2, 1> s_sum_acc_ct1(
-                sycl::range<1>(work_group_size / WARP_SIZE), cgh);
-
-            cgh.parallel_for(
-                sycl::nd_range<3>(sycl::range<3>(1, 1, nrows) * block_dims,
-                    block_dims),
-                [=](sycl::nd_item<3> item_ct1)
-                [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                    norm_f32(x, dst, ncols, eps, item_ct1,
-                        s_sum_acc_ct1.get_pointer(), work_group_size);
-                });
-            });
-    }
-}
-
-static void group_norm_f32_sycl(const float* x, float* dst,
-    const int num_groups, const int group_size,
-    const int ne_elements, queue_ptr stream) {
-    static const float eps = 1e-6f;
-    if (group_size < 1024) {
-        const sycl::range<3> block_dims(1, 1, WARP_SIZE);
-        stream->submit([&](sycl::handler& cgh) {
-            const float eps_ct4 = eps;
-            cgh.parallel_for(
-                sycl::nd_range<3>(sycl::range<3>(1, 1, num_groups) * block_dims,
-                    block_dims),
-                [=](sycl::nd_item<3> item_ct1)
-                [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                    group_norm_f32(
-                        x, dst, group_size, ne_elements, eps_ct4, item_ct1,
-                        nullptr, WARP_SIZE);
-                });
-            });
-    }
-    else {
-        const int work_group_size = get_work_group_size(stream->get_device());
-        const sycl::range<3> block_dims(1, 1, work_group_size);
-        /*
-        DPCT1049:18: The work-group size passed to the SYCL kernel may exceed
-        the limit. To get the device limit, query
-        info::device::max_work_group_size. Adjust the work-group size if needed.
-        */
-
-        stream->submit([&](sycl::handler& cgh) {
-            sycl::local_accessor<float, 1> s_sum_acc_ct1(sycl::range<1>(work_group_size / WARP_SIZE),
-                cgh);
-
-            const float eps_ct4 = eps;
-
-            cgh.parallel_for(
-                sycl::nd_range<3>(sycl::range<3>(1, 1, num_groups) * block_dims,
-                    block_dims),
-                [=](sycl::nd_item<3> item_ct1)
-                [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                    group_norm_f32(x, dst, group_size, ne_elements,
-                        eps_ct4, item_ct1,
-                        s_sum_acc_ct1.get_pointer(), work_group_size);
-                });
-            });
-    }
-}
-
-static void rms_norm_f32_sycl(const float* x, float* dst, const int ncols,
-    const int nrows, const float eps,
-    queue_ptr stream) {
-    GGML_ASSERT(ncols % WARP_SIZE == 0);
-    // printf("%s ncols=%d, nrows=%d, WARP_SIZE=%d\n", __func__, ncols, nrows, WARP_SIZE);
-    if (ncols < 1024) {
-        const sycl::range<3> block_dims(1, 1, WARP_SIZE);
-        stream->submit([&](sycl::handler& cgh) {
-            cgh.parallel_for(
-                sycl::nd_range<3>(sycl::range<3>(1, 1, nrows) * block_dims,
-                    block_dims),
-                [=](sycl::nd_item<3> item_ct1)
-                [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                    rms_norm_f32(x, dst, ncols, eps, item_ct1,
-                        nullptr, WARP_SIZE);
-                });
-            });
-    }
-    else {
-        const int work_group_size = get_work_group_size(stream->get_device());
-        const sycl::range<3> block_dims(1, 1, work_group_size);
-        /*
-        DPCT1049:19: The work-group size passed to the SYCL kernel may exceed
-        the limit. To get the device limit, query
-        info::device::max_work_group_size. Adjust the work-group size if needed.
-        */
-        stream->submit([&](sycl::handler& cgh) {
-            sycl::local_accessor<float, 1> s_sum_acc_ct1(sycl::range<1>(work_group_size / WARP_SIZE),
-                cgh);
-            cgh.parallel_for(
-                sycl::nd_range<3>(sycl::range<3>(1, 1, nrows) * block_dims,
-                    block_dims),
-                [=](sycl::nd_item<3> item_ct1)
-                [[intel::reqd_sub_group_size(WARP_SIZE)]] {
-                    rms_norm_f32(x, dst, ncols, eps, item_ct1,
-                        s_sum_acc_ct1.get_pointer(), work_group_size);
-                });
-            });
-    }
-}
-
-void ggml_sycl_op_norm(ggml_backend_sycl_context& ctx, const ggml_tensor* src0, const ggml_tensor* src1,
-    ggml_tensor* dst, const float* src0_dd,
-    const float* src1_dd, float* dst_dd,
-    const queue_ptr& main_stream) {
-
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT(dst->type == GGML_TYPE_F32);
-
-    const int64_t ne00 = src0->ne[0];
-    const int64_t nrows = ggml_nrows(src0);
-
-    float eps;
-    memcpy(&eps, dst->op_params, sizeof(float));
-
-    norm_f32_sycl(src0_dd, dst_dd, ne00, nrows, eps, main_stream);
-
-    (void)src1;
-    (void)dst;
-    (void)src1_dd;
-}
-
-void ggml_sycl_op_group_norm(ggml_backend_sycl_context& ctx, const ggml_tensor* src0,
-    const ggml_tensor* src1, ggml_tensor* dst,
-    const float* src0_dd, const float* src1_dd,
-    float* dst_dd,
-    const queue_ptr& main_stream) {
-
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT(dst->type == GGML_TYPE_F32);
-
-    int num_groups = dst->op_params[0];
-    int group_size = src0->ne[0] * src0->ne[1] * ((src0->ne[2] + num_groups - 1) / num_groups);
-    group_norm_f32_sycl(src0_dd, dst_dd, num_groups, group_size, src0->ne[0] * src0->ne[1] * src0->ne[2], main_stream);
-
-    (void)src1;
-    (void)dst;
-    (void)src1_dd;
-}
-
-void ggml_sycl_op_rms_norm(ggml_backend_sycl_context& ctx, const ggml_tensor* src0,
-    const ggml_tensor* src1, ggml_tensor* dst,
-    const float* src0_dd, const float* src1_dd,
-    float* dst_dd,
-    const queue_ptr& main_stream) {
-
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT(dst->type == GGML_TYPE_F32);
-
-    const int64_t ne00 = src0->ne[0];
-    const int64_t nrows = ggml_nrows(src0);
-
-    float eps;
-    memcpy(&eps, dst->op_params, sizeof(float));
-
-    rms_norm_f32_sycl(src0_dd, dst_dd, ne00, nrows, eps, main_stream);
-
-    (void)src1;
-    (void)dst;
-    (void)src1_dd;
-}
@@ -1,35 +0,0 @@
-//
-// MIT license
-// Copyright (C) 2024 Intel Corporation
-// SPDX-License-Identifier: MIT
-//
-
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-
-#ifndef GGML_SYCL_NORM_HPP
-#define GGML_SYCL_NORM_HPP
-
-#include "common.hpp"
-
-void ggml_sycl_op_norm(ggml_backend_sycl_context& ctx, const ggml_tensor* src0, const ggml_tensor* src1,
-    ggml_tensor* dst, const float* src0_dd,
-    const float* src1_dd, float* dst_dd,
-    const queue_ptr& main_stream);
-
-void ggml_sycl_op_rms_norm(ggml_backend_sycl_context& ctx, const ggml_tensor* src0,
-    const ggml_tensor* src1, ggml_tensor* dst,
-    const float* src0_dd, const float* src1_dd,
-    float* dst_dd,
-    const queue_ptr& main_stream);
-
-void ggml_sycl_op_group_norm(ggml_backend_sycl_context& ctx, const ggml_tensor* src0,
-    const ggml_tensor* src1, ggml_tensor* dst,
-    const float* src0_dd, const float* src1_dd,
-    float* dst_dd,
-    const queue_ptr& main_stream);
-
-#endif // GGML_SYCL_NORM_HPP
@@ -16,7 +16,7 @@
 #define GGML_SYCL_MAX_STREAMS       8
 #define GGML_SYCL_MAX_BUFFERS       256

-#define WARP_SIZE GGML_SYCL_WARP_SIZE
+#define WARP_SIZE 32
 #define MATRIX_ROW_PADDING 512 // last row of quant. matrices is a multiple of this to avoid out-of-bounds memory accesses

 #define SYCL_GELU_BLOCK_SIZE 256
@@ -1,275 +0,0 @@
-#include "rope.hpp"
-
-struct rope_corr_dims {
-    float v[2];
-};
-
-static float rope_yarn_ramp(const float low, const float high, const int i0) {
-    const float y = (i0 / 2 - low) / sycl::max(0.001f, high - low);
-    return 1.0f - sycl::min(1.0f, sycl::max(0.0f, y));
-}
-
-// YaRN algorithm based on LlamaYaRNScaledRotaryEmbedding.py from https://github.com/jquesnelle/yarn
-// MIT licensed. Copyright (c) 2023 Jeffrey Quesnelle and Bowen Peng.
-static void rope_yarn(
-    float theta_extrap, float freq_scale, rope_corr_dims corr_dims, int64_t i0, float ext_factor, float mscale,
-    float * cos_theta, float * sin_theta) {
-    // Get n-d rotational scaling corrected for extrapolation
-    float theta_interp = freq_scale * theta_extrap;
-    float theta = theta_interp;
-    if (ext_factor != 0.0f) {
-        float ramp_mix = rope_yarn_ramp(corr_dims.v[0], corr_dims.v[1], i0) * ext_factor;
-        theta = theta_interp * (1 - ramp_mix) + theta_extrap * ramp_mix;
-
-        // Get n-d magnitude scaling corrected for interpolation
-        mscale *= 1.0f + 0.1f * sycl::log(1.0f / freq_scale);
-    }
-    *cos_theta = sycl::cos(theta) * mscale;
-    *sin_theta = sycl::sin(theta) * mscale;
-}
-
-template<typename T, bool has_ff>
-static void rope_norm(
-    const T * x, T * dst, int ne0, int n_dims, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float ext_factor, float attn_factor, rope_corr_dims corr_dims, float theta_scale, const float * freq_factors,
-    const sycl::nd_item<3> &item_ct1) {
-    const int i0 = 2 * (item_ct1.get_local_range(1) * item_ct1.get_group(1) +
-                         item_ct1.get_local_id(1));
-
-    if (i0 >= ne0) {
-        return;
-    }
-
-    const int row = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
-                    item_ct1.get_local_id(2);
-
-    if (i0 >= n_dims) {
-        const int i = row*ne0 + i0;
-
-        dst[i + 0] = x[i + 0];
-        dst[i + 1] = x[i + 1];
-
-        return;
-    }
-
-    const int i = row*ne0 + i0;
-    const int i2 = row/p_delta_rows;
-
-    const float theta_base = pos[i2]*powf(theta_scale, i0/2.0f);
-
-    const float freq_factor = has_ff ? freq_factors[i0/2] : 1.0f;
-
-    float cos_theta;
-    float sin_theta;
-
-    rope_yarn(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, &cos_theta, &sin_theta);
-
-    const float x0 = x[i + 0];
-    const float x1 = x[i + 1];
-
-    dst[i + 0] = x0*cos_theta - x1*sin_theta;
-    dst[i + 1] = x0*sin_theta + x1*cos_theta;
-}
-
-template<typename T, bool has_ff>
-static void rope_neox(
-    const T * x, T * dst, int ne0, int n_dims, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float ext_factor, float attn_factor, rope_corr_dims corr_dims, float theta_scale, const float * freq_factors,
-    const sycl::nd_item<3> &item_ct1) {
-    const int i0 = 2 * (item_ct1.get_local_range(1) * item_ct1.get_group(1) +
-                         item_ct1.get_local_id(1));
-
-    if (i0 >= ne0) {
-        return;
-    }
-
-    const int row = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
-                    item_ct1.get_local_id(2);
-
-    if (i0 >= n_dims) {
-        const int i = row*ne0 + i0;
-
-        dst[i + 0] = x[i + 0];
-        dst[i + 1] = x[i + 1];
-
-        return;
-    }
-
-    const int i  = row*ne0 + i0/2;
-    const int i2 = row/p_delta_rows;
-
-    const float theta_base = pos[i2]*powf(theta_scale, i0/2.0f);
-
-    const float freq_factor = has_ff ? freq_factors[i0/2] : 1.0f;
-
-    float cos_theta;
-    float sin_theta;
-
-    rope_yarn(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, &cos_theta, &sin_theta);
-
-    const float x0 = x[i + 0];
-    const float x1 = x[i + n_dims/2];
-
-    dst[i + 0]        = x0*cos_theta - x1*sin_theta;
-    dst[i + n_dims/2] = x0*sin_theta + x1*cos_theta;
-}
-
-template <typename T>
-static void rope_norm_sycl(
-    const T *x, T *dst, int ne0, int n_dims, int nr, const int32_t *pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, queue_ptr stream) {
-    GGML_ASSERT(ne0 % 2 == 0);
-    const sycl::range<3> block_dims(1, SYCL_ROPE_BLOCK_SIZE, 1);
-    const int num_blocks_x = (ne0 + 2*SYCL_ROPE_BLOCK_SIZE - 1) / (2*SYCL_ROPE_BLOCK_SIZE);
-    const sycl::range<3> block_nums(1, num_blocks_x, nr);
-
-    const float theta_scale = powf(freq_base, -2.0f/n_dims);
-
-    dpct::has_capability_or_fail(stream->get_device(),
-                                     {sycl::aspect::fp16});
-
-    if (freq_factors == nullptr) {
-        /*
-        DPCT1049:40: The work-group size passed to the SYCL kernel may exceed
-        the limit. To get the device limit, query
-        info::device::max_work_group_size. Adjust the work-group size if needed.
-        */
-        stream->parallel_for(
-            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) {
-                rope_norm<T, false>(x, dst, ne0, n_dims, pos, freq_scale, p_delta_rows,
-                               ext_factor, attn_factor, corr_dims, theta_scale, freq_factors,
-                               item_ct1);
-            });
-    } else {
-        /*
-        DPCT1049:41: The work-group size passed to the SYCL kernel may exceed
-        the limit. To get the device limit, query
-        info::device::max_work_group_size. Adjust the work-group size if needed.
-        */
-        stream->parallel_for(
-            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) {
-                rope_norm<T, true>(x, dst, ne0, n_dims, pos, freq_scale, p_delta_rows,
-                              ext_factor, attn_factor, corr_dims, theta_scale, freq_factors,
-                              item_ct1);
-            });
-    }
-}
-
-template <typename T>
-static void rope_neox_sycl(
-    const T *x, T *dst, int ne0, int n_dims, int nr, const int32_t *pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, queue_ptr stream) {
-    GGML_ASSERT(ne0 % 2 == 0);
-    const sycl::range<3> block_dims(1, SYCL_ROPE_BLOCK_SIZE, 1);
-    const int num_blocks_x = (ne0 + 2*SYCL_ROPE_BLOCK_SIZE - 1) / (2*SYCL_ROPE_BLOCK_SIZE);
-    const sycl::range<3> block_nums(1, num_blocks_x, nr);
-
-    const float theta_scale = powf(freq_base, -2.0f/n_dims);
-
-    dpct::has_capability_or_fail(stream->get_device(),
-                                    {sycl::aspect::fp16});
-
-    if (freq_factors == nullptr) {
-        stream->parallel_for(
-            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) {
-                rope_neox<T, false>(x, dst, ne0, n_dims, pos, freq_scale,
-                                    p_delta_rows, ext_factor, attn_factor,
-                                    corr_dims, theta_scale, freq_factors,
-                                    item_ct1);
-            });
-    } else {
-        stream->parallel_for(
-            sycl::nd_range<3>(block_nums * block_dims, block_dims),
-            [=](sycl::nd_item<3> item_ct1) {
-                rope_neox<T, true>(x, dst, ne0, n_dims, pos, freq_scale,
-                                    p_delta_rows, ext_factor, attn_factor,
-                                    corr_dims, theta_scale, freq_factors,
-                                    item_ct1);
-            });
-    }
-}
-
-void ggml_sycl_op_rope(
-    ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1, ggml_tensor *dst,
-    const float *src0_dd, const float *src1_dd, float *dst_dd, const queue_ptr &main_stream) {
-    const ggml_tensor * src2 = dst->src[2];
-
-    GGML_ASSERT(src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_F16);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32 ||  dst->type == GGML_TYPE_F16);
-    GGML_ASSERT(src0->type == dst->type);
-
-    const int64_t ne00 = src0->ne[0];
-    const int64_t ne01 = src0->ne[1];
-    const int64_t nr = ggml_nrows(src0);
-
-    //const int n_past      = ((int32_t *) dst->op_params)[0];
-    const int n_dims      = ((int32_t *) dst->op_params)[1];
-    const int mode        = ((int32_t *) dst->op_params)[2];
-    //const int n_ctx       = ((int32_t *) dst->op_params)[3];
-    const int n_ctx_orig  = ((int32_t *) dst->op_params)[4];
-
-    // RoPE alteration for extended context
-    float freq_base;
-    float freq_scale;
-    float ext_factor;
-    float attn_factor;
-    float beta_fast;
-    float beta_slow;
-
-    memcpy(&freq_base,   (int32_t *) dst->op_params +  5, sizeof(float));
-    memcpy(&freq_scale,  (int32_t *) dst->op_params +  6, sizeof(float));
-    memcpy(&ext_factor,  (int32_t *) dst->op_params +  7, sizeof(float));
-    memcpy(&attn_factor, (int32_t *) dst->op_params +  8, sizeof(float));
-    memcpy(&beta_fast,   (int32_t *) dst->op_params +  9, sizeof(float));
-    memcpy(&beta_slow,   (int32_t *) dst->op_params + 10, sizeof(float));
-
-    const bool is_neox = mode & 2;
-
-    const int32_t * pos = (const int32_t *) src1_dd;
-
-    const float * freq_factors = nullptr;
-    if (src2 != nullptr) {
-        freq_factors = (const float *) src2->data;
-    }
-
-    rope_corr_dims corr_dims;
-    ggml_rope_yarn_corr_dims(n_dims, n_ctx_orig, freq_base, beta_fast, beta_slow, corr_dims.v);
-
-    // compute
-    if (is_neox) {
-        if (src0->type == GGML_TYPE_F32) {
-            rope_neox_sycl(
-                (const float *)src0_dd, (float *)dst_dd, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, main_stream
-            );
-        } else if (src0->type == GGML_TYPE_F16) {
-            rope_neox_sycl(
-                (const sycl::half *)src0_dd, (sycl::half *)dst_dd, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, main_stream
-            );
-        } else {
-            GGML_ASSERT(false);
-        }
-    } else {
-        if (src0->type == GGML_TYPE_F32) {
-            rope_norm_sycl(
-                (const float *)src0_dd, (float *)dst_dd, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, main_stream
-            );
-        } else if (src0->type == GGML_TYPE_F16) {
-            rope_norm_sycl(
-                (const sycl::half *)src0_dd, (sycl::half *)dst_dd, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, main_stream
-            );
-        } else {
-            GGML_ASSERT(false);
-        }
-    }
-
-    (void) src1;
-    (void) dst;
-    (void) src1_dd;
-}
@@ -1,22 +0,0 @@
-//
-// MIT license
-// Copyright (C) 2024 Intel Corporation
-// SPDX-License-Identifier: MIT
-//
-
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-
-#ifndef GGML_SYCL_ROPE_HPP
-#define GGML_SYCL_ROPE_HPP
-
-#include "common.hpp"
-
-void ggml_sycl_op_rope(
-    ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1, ggml_tensor *dst,
-    const float *src0_dd, const float *src1_dd, float *dst_dd, const queue_ptr &main_stream);
-
-#endif // GGML_SYCL_ROPE_HPP
@@ -820,6 +820,7 @@ vec_dot_iq2_xxs_q8_1(const void *__restrict__ vbq,
 #if QK_K == 256
    const block_iq2_xxs * bq2 = (const block_iq2_xxs *) vbq;

+#if QR2_XXS == 8
    const int ib32 = iqs;
    const uint16_t * q2 = bq2->qs + 4*ib32;
    const uint8_t  * aux8 = (const uint8_t *)q2;
@@ -837,6 +838,26 @@ vec_dot_iq2_xxs_q8_1(const void *__restrict__ vbq,
    }
    const float d = (float)bq2->d * (0.5f + aux32) * bq8_1[ib32].ds[0] * 0.25f;
    return d * sumi;
+#else
+    // iqs is 0...15
+    const int ib32 = iqs/2;
+    const int il = iqs%2;
+    const uint16_t * q2 = bq2->qs + 4*ib32;
+    const uint8_t  * aux8 = (const uint8_t *)q2;
+    const uint8_t  * grid1 = (const uint8_t *)(iq2xxs_grid + aux8[2*il+0]);
+    const uint8_t  * grid2 = (const uint8_t *)(iq2xxs_grid + aux8[2*il+1]);
+    const uint32_t aux32 = q2[2] | (q2[3] << 16);
+    const float d = (float)bq2->d * (0.5f + (aux32 >> 28)) * bq8_1[ib32].ds[0] * 0.25f;
+    const uint8_t signs1 = ksigns_iq2xs[(aux32 >> 14*il) & 127];
+    const uint8_t signs2 = ksigns_iq2xs[(aux32 >> (14*il + 7)) & 127];
+    const int8_t * q8 = bq8_1[ib32].qs + 16*il;
+    int sumi1 = 0, sumi2 = 0;
+    for (int j = 0; j < 8; ++j) {
+        sumi1 += q8[j+0] * grid1[j] * (signs1 & kmask_iq2xs[j] ? -1 : 1);
+        sumi2 += q8[j+8] * grid2[j] * (signs2 & kmask_iq2xs[j] ? -1 : 1);
+    }
+    return d * (sumi1 + sumi2);
+#endif
 #else
    assert(false);
    return 0.f;
@@ -144954,3 +144954,4 @@ unsigned char sum_rows_f32_data[] = {

 };
 const uint64_t sum_rows_f32_len = 2112;
+
@@ -50,8 +50,6 @@ class Keys:
        POOLING_TYPE                      = "{arch}.pooling_type"
        LOGIT_SCALE                       = "{arch}.logit_scale"
        DECODER_START_TOKEN_ID            = "{arch}.decoder_start_token_id"
-        ATTN_LOGIT_SOFTCAPPING            = "{arch}.attn_logit_softcapping"
-        FINAL_LOGIT_SOFTCAPPING           = "{arch}.final_logit_softcapping"

    class Attention:
        HEAD_COUNT        = "{arch}.attention.head_count"
@@ -66,7 +64,6 @@ class Keys:
        Q_LORA_RANK       = "{arch}.attention.q_lora_rank"
        KV_LORA_RANK      = "{arch}.attention.kv_lora_rank"
        REL_BUCKETS_COUNT = "{arch}.attention.relative_buckets_count"
-        SLIDING_WINDOW    = "{arch}.attention.sliding_window"

    class Rope:
        DIMENSION_COUNT         = "{arch}.rope.dimension_count"
@@ -153,7 +150,6 @@ class MODEL_ARCH(IntEnum):
    INTERNLM2    = auto()
    MINICPM      = auto()
    GEMMA        = auto()
-    GEMMA2       = auto()
    STARCODER2   = auto()
    MAMBA        = auto()
    XVERSE       = auto()
@@ -164,7 +160,6 @@ class MODEL_ARCH(IntEnum):
    DEEPSEEK2    = auto()
    BITNET       = auto()
    T5           = auto()
-    JAIS         = auto()


 class MODEL_TENSOR(IntEnum):
@@ -185,13 +180,10 @@ class MODEL_TENSOR(IntEnum):
    ATTN_NORM            = auto()
    ATTN_NORM_2          = auto()
    ATTN_OUT_NORM        = auto()
-    ATTN_POST_NORM       = auto()
    ATTN_ROT_EMBD        = auto()
    FFN_GATE_INP         = auto()
    FFN_GATE_INP_SHEXP   = auto()
    FFN_NORM             = auto()
-    FFN_PRE_NORM         = auto()
-    FFN_POST_NORM        = auto()
    FFN_GATE             = auto()
    FFN_DOWN             = auto()
    FFN_UP               = auto()
@@ -278,7 +270,6 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
    MODEL_ARCH.INTERNLM2:      "internlm2",
    MODEL_ARCH.MINICPM:        "minicpm",
    MODEL_ARCH.GEMMA:          "gemma",
-    MODEL_ARCH.GEMMA2:         "gemma2",
    MODEL_ARCH.STARCODER2:     "starcoder2",
    MODEL_ARCH.MAMBA:          "mamba",
    MODEL_ARCH.XVERSE:         "xverse",
@@ -289,7 +280,6 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
    MODEL_ARCH.DEEPSEEK2:      "deepseek2",
    MODEL_ARCH.BITNET:         "bitnet",
    MODEL_ARCH.T5:             "t5",
-    MODEL_ARCH.JAIS:           "jais",
 }

 TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@@ -313,12 +303,9 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
    MODEL_TENSOR.ATTN_Q_NORM:          "blk.{bid}.attn_q_norm",
    MODEL_TENSOR.ATTN_K_NORM:          "blk.{bid}.attn_k_norm",
    MODEL_TENSOR.ATTN_OUT_NORM:        "blk.{bid}.attn_output_norm",
-    MODEL_TENSOR.ATTN_POST_NORM:       "blk.{bid}.post_attention_norm",
    MODEL_TENSOR.FFN_GATE_INP:         "blk.{bid}.ffn_gate_inp",
    MODEL_TENSOR.FFN_GATE_INP_SHEXP:   "blk.{bid}.ffn_gate_inp_shexp",
    MODEL_TENSOR.FFN_NORM:             "blk.{bid}.ffn_norm",
-    MODEL_TENSOR.FFN_PRE_NORM:         "blk.{bid}.ffn_norm",
-    MODEL_TENSOR.FFN_POST_NORM:        "blk.{bid}.post_ffw_norm",
    MODEL_TENSOR.FFN_GATE:             "blk.{bid}.ffn_gate",
    MODEL_TENSOR.FFN_DOWN:             "blk.{bid}.ffn_down",
    MODEL_TENSOR.FFN_UP:               "blk.{bid}.ffn_up",
@@ -764,21 +751,6 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
        MODEL_TENSOR.FFN_UP,
        MODEL_TENSOR.FFN_NORM,
    ],
-    MODEL_ARCH.GEMMA2: [
-        MODEL_TENSOR.TOKEN_EMBD,
-        MODEL_TENSOR.OUTPUT_NORM,
-        MODEL_TENSOR.ATTN_Q,
-        MODEL_TENSOR.ATTN_K,
-        MODEL_TENSOR.ATTN_V,
-        MODEL_TENSOR.ATTN_OUT,
-        MODEL_TENSOR.FFN_GATE,
-        MODEL_TENSOR.FFN_DOWN,
-        MODEL_TENSOR.FFN_UP,
-        MODEL_TENSOR.ATTN_NORM,
-        MODEL_TENSOR.ATTN_POST_NORM,
-        MODEL_TENSOR.FFN_PRE_NORM,
-        MODEL_TENSOR.FFN_POST_NORM,
-    ],
    MODEL_ARCH.STARCODER2: [
        MODEL_TENSOR.TOKEN_EMBD,
        MODEL_TENSOR.OUTPUT_NORM,
@@ -956,18 +928,6 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
        MODEL_TENSOR.ENC_FFN_UP,
        MODEL_TENSOR.ENC_OUTPUT_NORM,
    ],
-    MODEL_ARCH.JAIS: [
-        MODEL_TENSOR.TOKEN_EMBD,
-        MODEL_TENSOR.OUTPUT_NORM,
-        MODEL_TENSOR.OUTPUT,
-        MODEL_TENSOR.ATTN_NORM,
-        MODEL_TENSOR.ATTN_QKV,
-        MODEL_TENSOR.ATTN_OUT,
-        MODEL_TENSOR.FFN_NORM,
-        MODEL_TENSOR.FFN_DOWN,
-        MODEL_TENSOR.FFN_GATE,
-        MODEL_TENSOR.FFN_UP,
-    ],
    # TODO
 }

@@ -516,12 +516,6 @@ class GGUFWriter:
    def add_logit_scale(self, value: float) -> None:
        self.add_float32(Keys.LLM.LOGIT_SCALE.format(arch=self.arch), value)

-    def add_attn_logit_softcapping(self, value: float) -> None:
-        self.add_float32(Keys.LLM.ATTN_LOGIT_SOFTCAPPING.format(arch=self.arch), value)
-
-    def add_final_logit_softcapping(self, value: float) -> None:
-        self.add_float32(Keys.LLM.FINAL_LOGIT_SOFTCAPPING.format(arch=self.arch), value)
-
    def add_expert_count(self, count: int) -> None:
        self.add_uint32(Keys.LLM.EXPERT_COUNT.format(arch=self.arch), count)

@@ -552,9 +546,6 @@ class GGUFWriter:
    def add_relative_attn_buckets_count(self, value: int) -> None:
        self.add_uint32(Keys.Attention.REL_BUCKETS_COUNT.format(arch=self.arch), value)

-    def add_sliding_window(self, value: int) -> None:
-        self.add_uint32(Keys.Attention.SLIDING_WINDOW.format(arch=self.arch), value)
-
    def add_pooling_type(self, value: PoolingType) -> None:
        self.add_uint32(Keys.LLM.POOLING_TYPE.format(arch=self.arch), value.value)

@@ -10,7 +10,7 @@ class TensorNameMap:
        # Token embeddings
        MODEL_TENSOR.TOKEN_EMBD: (
            "gpt_neox.embed_in",                         # gptneox
-            "transformer.wte",                           # gpt2 gpt-j mpt refact qwen dbrx jais
+            "transformer.wte",                           # gpt2 gpt-j mpt refact qwen dbrx
            "transformer.word_embeddings",               # falcon
            "word_embeddings",                           # bloom
            "model.embed_tokens",                        # llama-hf
@@ -49,7 +49,7 @@ class TensorNameMap:
        # Output
        MODEL_TENSOR.OUTPUT: (
            "embed_out",                 # gptneox
-            "lm_head",                   # gpt2 mpt falcon llama-hf baichuan qwen mamba dbrx jais
+            "lm_head",                   # gpt2 mpt falcon llama-hf baichuan qwen mamba dbrx
            "output",                    # llama-pth bloom internlm2
            "word_embeddings_for_head",  # persimmon
            "lm_head.linear",            # phi2
@@ -58,7 +58,7 @@ class TensorNameMap:
        # Output norm
        MODEL_TENSOR.OUTPUT_NORM: (
            "gpt_neox.final_layer_norm",               # gptneox
-            "transformer.ln_f",                        # gpt2 gpt-j falcon jais
+            "transformer.ln_f",                        # gpt2 gpt-j falcon
            "model.norm",                              # llama-hf baichuan internlm2
            "norm",                                    # llama-pth
            "transformer.norm_f",                      # mpt dbrx
@@ -81,7 +81,7 @@ class TensorNameMap:
        # Attention norm
        MODEL_TENSOR.ATTN_NORM: (
            "gpt_neox.layers.{bid}.input_layernorm",                # gptneox
-            "transformer.h.{bid}.ln_1",                             # gpt2 gpt-j refact qwen jais
+            "transformer.h.{bid}.ln_1",                             # gpt2 gpt-j refact qwen
            "transformer.blocks.{bid}.norm_1",                      # mpt
            "transformer.h.{bid}.input_layernorm",                  # falcon7b
            "h.{bid}.input_layernorm",                              # bloom
@@ -109,7 +109,7 @@ class TensorNameMap:
        # Attention query-key-value
        MODEL_TENSOR.ATTN_QKV: (
            "gpt_neox.layers.{bid}.attention.query_key_value",                     # gptneox
-            "transformer.h.{bid}.attn.c_attn",                                     # gpt2 qwen jais
+            "transformer.h.{bid}.attn.c_attn",                                     # gpt2 qwen
            "transformer.blocks.{bid}.attn.Wqkv",                                  # mpt
            "transformer.blocks.{bid}.norm_attn_norm.attn.Wqkv",                   # dbrx
            "transformer.h.{bid}.self_attention.query_key_value",                  # falcon
@@ -160,7 +160,7 @@ class TensorNameMap:
        # Attention output
        MODEL_TENSOR.ATTN_OUT: (
            "gpt_neox.layers.{bid}.attention.dense",                        # gptneox
-            "transformer.h.{bid}.attn.c_proj",                              # gpt2 refact qwen jais
+            "transformer.h.{bid}.attn.c_proj",                              # gpt2 refact qwen
            "transformer.blocks.{bid}.attn.out_proj",                       # mpt
            "transformer.h.{bid}.self_attention.dense",                     # falcon
            "h.{bid}.self_attention.dense",                                 # bloom
@@ -187,10 +187,6 @@ class TensorNameMap:
            "transformer.blocks.{bid}.norm_attn_norm.norm_2",  # dbrx
        ),

-        MODEL_TENSOR.ATTN_POST_NORM: (
-            "model.layers.{bid}.post_attention_layernorm",     # gemma2
-        ),
-
        # Rotary embeddings
        MODEL_TENSOR.ATTN_ROT_EMBD: (
            "model.layers.{bid}.self_attn.rotary_emb.inv_freq",        # llama-hf
@@ -202,7 +198,7 @@ class TensorNameMap:
        # Feed-forward norm
        MODEL_TENSOR.FFN_NORM: (
            "gpt_neox.layers.{bid}.post_attention_layernorm",                # gptneox
-            "transformer.h.{bid}.ln_2",                                      # gpt2 refact qwen jais
+            "transformer.h.{bid}.ln_2",                                      # gpt2 refact qwen
            "h.{bid}.post_attention_layernorm",                              # bloom
            "transformer.blocks.{bid}.norm_2",                               # mpt
            "model.layers.{bid}.post_attention_layernorm",                   # llama-hf
@@ -214,16 +210,6 @@ class TensorNameMap:
            "transformer.decoder_layer.{bid}.rms_norm_2",                    # Grok
        ),

-        # Post feed-forward norm
-        MODEL_TENSOR.FFN_PRE_NORM: (
-            "model.layers.{bid}.pre_feedforward_layernorm", # gemma2
-        ),
-
-        # Post feed-forward norm
-        MODEL_TENSOR.FFN_POST_NORM: (
-            "model.layers.{bid}.post_feedforward_layernorm", # gemma2
-        ),
-
        MODEL_TENSOR.FFN_GATE_INP: (
            "layers.{bid}.feed_forward.gate",             # mixtral
            "model.layers.{bid}.block_sparse_moe.gate",   # mixtral
@@ -239,7 +225,7 @@ class TensorNameMap:
        # Feed-forward up
        MODEL_TENSOR.FFN_UP: (
            "gpt_neox.layers.{bid}.mlp.dense_h_to_4h",                # gptneox
-            "transformer.h.{bid}.mlp.c_fc",                           # gpt2 jais
+            "transformer.h.{bid}.mlp.c_fc",                           # gpt2
            "transformer.blocks.{bid}.ffn.up_proj",                   # mpt
            "transformer.h.{bid}.mlp.dense_h_to_4h",                  # falcon
            "h.{bid}.mlp.dense_h_to_4h",                              # bloom
@@ -285,7 +271,6 @@ class TensorNameMap:
            "model.layers.{bid}.mlp.gate_proj",           # llama-hf refact
            "layers.{bid}.feed_forward.w1",               # llama-pth
            "transformer.h.{bid}.mlp.w2",                 # qwen
-            "transformer.h.{bid}.mlp.c_fc2",              # jais
            "model.layers.layers.{bid}.mlp.gate_proj",    # plamo
            "model.layers.{bid}.feed_forward.w1",         # internlm2
            "encoder.layers.{bid}.mlp.fc12",              # nomic-bert
@@ -309,7 +294,7 @@ class TensorNameMap:
        # Feed-forward down
        MODEL_TENSOR.FFN_DOWN: (
            "gpt_neox.layers.{bid}.mlp.dense_4h_to_h",                # gptneox
-            "transformer.h.{bid}.mlp.c_proj",                         # gpt2 refact qwen jais
+            "transformer.h.{bid}.mlp.c_proj",                         # gpt2 refact qwen
            "transformer.blocks.{bid}.ffn.down_proj",                 # mpt
            "transformer.h.{bid}.mlp.dense_4h_to_h",                  # falcon
            "h.{bid}.mlp.dense_4h_to_h",                              # bloom
@@ -126,257 +126,19 @@ You can use GBNF grammars:
    - in CLI, with [examples/json_schema_to_grammar.py](../examples/json_schema_to_grammar.py)
    - in JavaScript with [json-schema-to-grammar.mjs](../examples/server/public/json-schema-to-grammar.mjs) (this is used by the [server](../examples/server)'s Web UI)

-Take a look at [tests](../tests/test-json-schema-to-grammar.cpp) to see which features are likely supported (you'll also find usage examples in https://github.com/ggerganov/llama.cpp/pull/5978, https://github.com/ggerganov/llama.cpp/pull/6659 & https://github.com/ggerganov/llama.cpp/pull/6555).
+Take a look at [tests](../../tests/test-json-schema-to-grammar.cpp) to see which features are likely supported (you'll also find usage examples in https://github.com/ggerganov/llama.cpp/pull/5978, https://github.com/ggerganov/llama.cpp/pull/6659 & https://github.com/ggerganov/llama.cpp/pull/6555).

-```bash
-llama-cli \
-  -hfr bartowski/Phi-3-medium-128k-instruct-GGUF \
-  -hff Phi-3-medium-128k-instruct-Q8_0.gguf \
-  -j '{
-    "type": "array",
-    "items": {
-        "type": "object",
-        "properties": {
-            "name": {
-                "type": "string",
-                "minLength": 1,
-                "maxLength": 100
-            },
-            "age": {
-                "type": "integer",
-                "minimum": 0,
-                "maximum": 150
-            }
-        },
-        "required": ["name", "age"],
-        "additionalProperties": false
-    },
-    "minItems": 10,
-    "maxItems": 100
-  }' \
-  -p 'Generate a {name, age}[] JSON array with famous actors of all ages.'
-```
+Here is also a non-exhaustive list of **unsupported** features:

-<details>
-
-<summary>Show grammar</summary>
-
-You can convert any schema in command-line with:
-
-```bash
-examples/json_schema_to_grammar.py name-age-schema.json
-```
-
-```
-char ::= [^"\\\x7F\x00-\x1F] | [\\] (["\\bfnrt] | "u" [0-9a-fA-F]{4})
-item ::= "{" space item-name-kv "," space item-age-kv "}" space
-item-age ::= ([0-9] | ([1-8] [0-9] | [9] [0-9]) | "1" ([0-4] [0-9] | [5] "0")) space
-item-age-kv ::= "\"age\"" space ":" space item-age
-item-name ::= "\"" char{1,100} "\"" space
-item-name-kv ::= "\"name\"" space ":" space item-name
-root ::= "[" space item ("," space item){9,99} "]" space
-space ::= | " " | "\n" [ \t]{0,20}
-```
-
-</details>
-
-Here is also a list of known limitations (contributions welcome):
-
- `additionalProperties` defaults to `false` (produces faster grammars + reduces hallucinations).
- `"additionalProperties": true` may produce keys that contain unescaped newlines.
- Unsupported features are skipped silently. It is currently advised to use the command-line Python converter (see above) to see any warnings, and to inspect the resulting grammar / test it w/ [llama-gbnf-validator](../examples/gbnf-validator/gbnf-validator.cpp).
- Can't mix `properties` w/ `anyOf` / `oneOf` in the same type (https://github.com/ggerganov/llama.cpp/issues/7703)
- [prefixItems](https://json-schema.org/draft/2020-12/json-schema-core#name-prefixitems) is broken (but [items](https://json-schema.org/draft/2020-12/json-schema-core#name-items) works)
- `minimum`, `exclusiveMinimum`, `maximum`, `exclusiveMaximum`: only supported for `"type": "integer"` for now, not `number`
- Nested `$ref`s are broken (https://github.com/ggerganov/llama.cpp/issues/8073)
- [pattern](https://json-schema.org/draft/2020-12/json-schema-validation#name-pattern)s must start with `^` and end with `$`
- Remote `$ref`s not supported in the C++ version (Python & JavaScript versions fetch https refs)
- `string` [formats](https://json-schema.org/draft/2020-12/json-schema-validation#name-defined-formats) lack `uri`, `email`
- No [`patternProperties`](https://json-schema.org/draft/2020-12/json-schema-core#name-patternproperties)
-
-And a non-exhaustive list of other unsupported features that are unlikely to be implemented (hard and/or too slow to support w/ stateless grammars):
-
- [`uniqueItems`](https://json-schema.org/draft/2020-12/json-schema-validation#name-uniqueitems)
+- `additionalProperties`: to be fixed in https://github.com/ggerganov/llama.cpp/pull/7840
+- `minimum`, `exclusiveMinimum`, `maximum`, `exclusiveMaximum`
+    - `integer` constraints to be implemented in https://github.com/ggerganov/llama.cpp/pull/7797
+- Remote `$ref`s in the C++ version (Python & JavaScript versions fetch https refs)
+- Mixing `properties` w/ `anyOf` / `oneOf` in the same type (https://github.com/ggerganov/llama.cpp/issues/7703)
+- `string` formats `uri`, `email`
 - [`contains`](https://json-schema.org/draft/2020-12/json-schema-core#name-contains) / `minContains`
+- `uniqueItems`
 - `$anchor` (cf. [dereferencing](https://json-schema.org/draft/2020-12/json-schema-core#name-dereferencing))
 - [`not`](https://json-schema.org/draft/2020-12/json-schema-core#name-not)
 - [Conditionals](https://json-schema.org/draft/2020-12/json-schema-core#name-keywords-for-applying-subsche) `if` / `then` / `else` / `dependentSchemas`
-
-### A word about additionalProperties
-
-> [!WARNING]
-> The JSON schemas spec states `object`s accept [additional properties](https://json-schema.org/understanding-json-schema/reference/object#additionalproperties) by default.
-> Since this is slow and seems prone to hallucinations, we default to no additional properties.
-> You can set `"additionalProperties": true` in the the schema of any object to explicitly allow additional properties.
-
-If you're using [Pydantic](https://pydantic.dev/) to generate schemas, you can enable additional properties with the `extra` config on each model class:
-
-```python
-# pip install pydantic
-import json
-from typing import Annotated, List
-from pydantic import BaseModel, Extra, Field
-class QAPair(BaseModel):
-    class Config:
-        extra = 'allow'  # triggers additionalProperties: true in the JSON schema
-    question: str
-    concise_answer: str
-    justification: str
-
-class Summary(BaseModel):
-    class Config:
-        extra = 'allow'
-    key_facts: List[Annotated[str, Field(pattern='- .{5,}')]]
-    question_answers: List[Annotated[List[QAPair], Field(min_items=5)]]
-
-print(json.dumps(Summary.model_json_schema(), indent=2))
-```
-
-<details>
-<summary>Show JSON schema & grammar</summary>
-
-```json
-{
-  "$defs": {
-    "QAPair": {
-      "additionalProperties": true,
-      "properties": {
-        "question": {
-          "title": "Question",
-          "type": "string"
-        },
-        "concise_answer": {
-          "title": "Concise Answer",
-          "type": "string"
-        },
-        "justification": {
-          "title": "Justification",
-          "type": "string"
-        }
-      },
-      "required": [
-        "question",
-        "concise_answer",
-        "justification"
-      ],
-      "title": "QAPair",
-      "type": "object"
-    }
-  },
-  "additionalProperties": true,
-  "properties": {
-    "key_facts": {
-      "items": {
-        "pattern": "^- .{5,}$",
-        "type": "string"
-      },
-      "title": "Key Facts",
-      "type": "array"
-    },
-    "question_answers": {
-      "items": {
-        "items": {
-          "$ref": "#/$defs/QAPair"
-        },
-        "minItems": 5,
-        "type": "array"
-      },
-      "title": "Question Answers",
-      "type": "array"
-    }
-  },
-  "required": [
-    "key_facts",
-    "question_answers"
-  ],
-  "title": "Summary",
-  "type": "object"
-}
-```
-
-```
-QAPair ::= "{" space QAPair-question-kv "," space QAPair-concise-answer-kv "," space QAPair-justification-kv ( "," space ( QAPair-additional-kv ( "," space QAPair-additional-kv )* ) )? "}" space
-QAPair-additional-k ::= ["] ( [c] ([o] ([n] ([c] ([i] ([s] ([e] ([_] ([a] ([n] ([s] ([w] ([e] ([r] char+ | [^"r] char*) | [^"e] char*) | [^"w] char*) | [^"s] char*) | [^"n] char*) | [^"a] char*) | [^"_] char*) | [^"e] char*) | [^"s] char*) | [^"i] char*) | [^"c] char*) | [^"n] char*) | [^"o] char*) | [j] ([u] ([s] ([t] ([i] ([f] ([i] ([c] ([a] ([t] ([i] ([o] ([n] char+ | [^"n] char*) | [^"o] char*) | [^"i] char*) | [^"t] char*) | [^"a] char*) | [^"c] char*) | [^"i] char*) | [^"f] char*) | [^"i] char*) | [^"t] char*) | [^"s] char*) | [^"u] char*) | [q] ([u] ([e] ([s] ([t] ([i] ([o] ([n] char+ | [^"n] char*) | [^"o] char*) | [^"i] char*) | [^"t] char*) | [^"s] char*) | [^"e] char*) | [^"u] char*) | [^"cjq] char* )? ["] space
-QAPair-additional-kv ::= QAPair-additional-k ":" space value
-QAPair-concise-answer-kv ::= "\"concise_answer\"" space ":" space string
-QAPair-justification-kv ::= "\"justification\"" space ":" space string
-QAPair-question-kv ::= "\"question\"" space ":" space string
-additional-k ::= ["] ( [k] ([e] ([y] ([_] ([f] ([a] ([c] ([t] ([s] char+ | [^"s] char*) | [^"t] char*) | [^"c] char*) | [^"a] char*) | [^"f] char*) | [^"_] char*) | [^"y] char*) | [^"e] char*) | [q] ([u] ([e] ([s] ([t] ([i] ([o] ([n] ([_] ([a] ([n] ([s] ([w] ([e] ([r] ([s] char+ | [^"s] char*) | [^"r] char*) | [^"e] char*) | [^"w] char*) | [^"s] char*) | [^"n] char*) | [^"a] char*) | [^"_] char*) | [^"n] char*) | [^"o] char*) | [^"i] char*) | [^"t] char*) | [^"s] char*) | [^"e] char*) | [^"u] char*) | [^"kq] char* )? ["] space
-additional-kv ::= additional-k ":" space value
-array ::= "[" space ( value ("," space value)* )? "]" space
-boolean ::= ("true" | "false") space
-char ::= [^"\\\x7F\x00-\x1F] | [\\] (["\\bfnrt] | "u" [0-9a-fA-F]{4})
-decimal-part ::= [0-9]{1,16}
-dot ::= [^\x0A\x0D]
-integral-part ::= [0] | [1-9] [0-9]{0,15}
-key-facts ::= "[" space (key-facts-item ("," space key-facts-item)*)? "]" space
-key-facts-item ::= "\"" "- " key-facts-item-1{5,} "\"" space
-key-facts-item-1 ::= dot
-key-facts-kv ::= "\"key_facts\"" space ":" space key-facts
-null ::= "null" space
-number ::= ("-"? integral-part) ("." decimal-part)? ([eE] [-+]? integral-part)? space
-object ::= "{" space ( string ":" space value ("," space string ":" space value)* )? "}" space
-question-answers ::= "[" space (question-answers-item ("," space question-answers-item)*)? "]" space
-question-answers-item ::= "[" space question-answers-item-item ("," space question-answers-item-item){4,} "]" space
-question-answers-item-item ::= QAPair
-question-answers-kv ::= "\"question_answers\"" space ":" space question-answers
-root ::= "{" space key-facts-kv "," space question-answers-kv ( "," space ( additional-kv ( "," space additional-kv )* ) )? "}" space
-space ::= | " " | "\n" [ \t]{0,20}
-string ::= "\"" char* "\"" space
-value ::= object | array | string | number | boolean | null
-```
-
-</details>
-
-If you're using [Zod](https://zod.dev/), you can make your objects to explicitly allow extra properties w/ `nonstrict()` / `passthrough()` (or explicitly no extra props w/ `z.object(...).strict()` or `z.strictObject(...)`) but note that [zod-to-json-schema](https://github.com/StefanTerdell/zod-to-json-schema) currently always sets `"additionalProperties": false` anyway.
-
-```js
-import { z } from 'zod';
-import { zodToJsonSchema } from 'zod-to-json-schema';
-
-const Foo = z.object({
-  age: z.number().positive(),
-  email: z.string().email(),
-}).strict();
-
-console.log(zodToJsonSchema(Foo));
-```
-
-<details>
-<summary>Show JSON schema & grammar</summary>
-
-```json
-{
-  "type": "object",
-  "properties": {
-    "age": {
-      "type": "number",
-      "exclusiveMinimum": 0
-    },
-    "email": {
-      "type": "string",
-      "format": "email"
-    }
-  },
-  "required": [
-    "age",
-    "email"
-  ],
-  "additionalProperties": false,
-  "$schema": "http://json-schema.org/draft-07/schema#"
-}
-```
-
-```
-age-kv ::= "\"age\"" space ":" space number
-char ::= [^"\\\x7F\x00-\x1F] | [\\] (["\\bfnrt] | "u" [0-9a-fA-F]{4})
-decimal-part ::= [0-9]{1,16}
-email-kv ::= "\"email\"" space ":" space string
-integral-part ::= [0] | [1-9] [0-9]{0,15}
-number ::= ("-"? integral-part) ("." decimal-part)? ([eE] [-+]? integral-part)? space
-root ::= "{" space age-kv "," space email-kv "}" space
-space ::= | " " | "\n" [ \t]{0,20}
-string ::= "\"" char* "\"" space
-```
-
-</details>
+- [`patternProperties`](https://json-schema.org/draft/2020-12/json-schema-core#name-patternproperties)
@@ -88,8 +88,6 @@ extern "C" {
        LLAMA_VOCAB_PRE_TYPE_DBRX           = 13,
        LLAMA_VOCAB_PRE_TYPE_SMAUG          = 14,
        LLAMA_VOCAB_PRE_TYPE_PORO           = 15,
-        LLAMA_VOCAB_PRE_TYPE_VIKING         = 16,
-        LLAMA_VOCAB_PRE_TYPE_JAIS           = 17,
    };

    // note: these values should be synchronized with ggml_rope
@@ -210,3 +210,4 @@ fi
 # more benches
 #GGML_CUDA=1 make -j && ./llama-batched-bench ./models/codellama-7b/ggml-model-q4_k.gguf  4096 1 99 1 512,3200 128,128,800 1
 #GGML_CUDA=1 make -j && ./llama-batched-bench ./models/codellama-13b/ggml-model-q4_k.gguf 4096 1 99 1 512,3200 128,128,800 1
+
@@ -136,41 +136,42 @@ if [ -f $SRC_LLAMA/ggml-src.patch ]; then
    # LICENSE                -> LICENSE
    # scripts/gen-authors.sh -> scripts/gen-authors.sh

-    cat ggml-src.patch | sed -E \
-        -e 's/([[:space:]]|[ab]\/)CMakeLists.txt/\1ggml\/CMakeLists.txt/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/CMakeLists.txt/\1ggml\/src\/CMakeLists.txt/g' \
-        -e 's/([[:space:]]|[ab]\/)cmake\/FindSIMD.cmake/\1ggml\/cmake\/FindSIMD.cmake/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml\.c/\1ggml\/src\/ggml.c/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-alloc\.c/\1ggml\/src\/ggml-alloc.c/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-backend-impl\.h/\1ggml\/src\/ggml-backend-impl.h/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-backend\.c/\1ggml\/src\/ggml-backend.c/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-common\.h/\1ggml\/src\/ggml-common.h/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-cuda\//\1ggml\/src\/ggml-cuda\//g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-cuda\.cu/\1ggml\/src\/ggml-cuda.cu/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-impl\.h/\1ggml\/src\/ggml-impl.h/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-kompute\.cpp/\1ggml\/src\/ggml-kompute.cpp/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-metal\.m/\1ggml\/src\/ggml-metal.m/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-quants\.c/\1ggml\/src\/ggml-quants.c/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-quants\.h/\1ggml\/src\/ggml-quants.h/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-rpc\.cpp/\1ggml\/src\/ggml-rpc.cpp/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-sycl\.cpp/\1ggml\/src\/ggml-sycl.cpp/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-vulkan\.cpp/\1ggml\/src\/ggml-vulkan.cpp/g' \
-        -e 's/([[:space:]]|[ab]\/)include\/ggml\.h/\1ggml\/include\/ggml.h/g' \
-        -e 's/([[:space:]]|[ab]\/)include\/ggml-alloc\.h/\1ggml\/include\/ggml-alloc.h/g' \
-        -e 's/([[:space:]]|[ab]\/)include\/ggml-backend\.h/\1ggml\/include\/ggml-backend.h/g' \
-        -e 's/([[:space:]]|[ab]\/)include\/ggml-blas\.h/\1ggml\/include\/ggml-blas.h/g' \
-        -e 's/([[:space:]]|[ab]\/)include\/ggml-cuda\.h/\1ggml\/include\/ggml-cuda.h/g' \
-        -e 's/([[:space:]]|[ab]\/)include\/ggml-kompute\.h/\1ggml\/include\/ggml-kompute.h/g' \
-        -e 's/([[:space:]]|[ab]\/)include\/ggml-metal\.h/\1ggml\/include\/ggml-metal.h/g' \
-        -e 's/([[:space:]]|[ab]\/)include\/ggml-rpc\.h/\1ggml\/include\/ggml-rpc.h/g' \
-        -e 's/([[:space:]]|[ab]\/)include\/ggml-sycl\.h/\1ggml\/include\/ggml-sycl.h/g' \
-        -e 's/([[:space:]]|[ab]\/)include\/ggml-vulkan\.h/\1ggml\/include\/ggml-vulkan.h/g' \
-        -e 's/([[:space:]]|[ab]\/)examples\/common\.h/examples\/common.h/g' \
-        -e 's/([[:space:]]|[ab]\/)examples\/common\.cpp/examples\/common.cpp/g' \
-        -e 's/([[:space:]]|[ab]\/)examples\/common-ggml\.h/examples\/common-ggml.h/g' \
-        -e 's/([[:space:]]|[ab]\/)examples\/common-ggml\.cpp/examples\/common-ggml.cpp/g' \
-        -e 's/([[:space:]]|[ab]\/)LICENSE/LICENSE/g' \
-        -e 's/([[:space:]]|[ab]\/)scripts\/gen-authors\.sh/scripts\/gen-authors.sh/g' \
+    cat ggml-src.patch | sed \
+        -e 's/CMakeLists.txt/ggml\/CMakeLists.txt/g' \
+        -e 's/src\/CMakeLists.txt/ggml\/src\/CMakeLists.txt/g' \
+        -e 's/cmake\/FindSIMD.cmake/ggml\/cmake\/FindSIMD.cmake/g' \
+        -e 's/src\/ggml\.c/ggml/src/ggml.c/g' \
+        -e 's/src\/ggml-alloc\.c/ggml/src/ggml-alloc.c/g' \
+        -e 's/src\/ggml-backend-impl\.h/ggml/src/ggml-backend-impl.h/g' \
+        -e 's/src\/ggml-backend\.c/ggml/src/ggml-backend.c/g' \
+        -e 's/src\/ggml-common\.h/ggml/src/ggml-common.h/g' \
+        -e 's/src\/ggml-cuda\//ggml-cuda\//g' \
+        -e 's/src\/ggml-cuda\.cu/ggml/src/ggml-cuda.cu/g' \
+        -e 's/src\/ggml-impl\.h/ggml/src/ggml-impl.h/g' \
+        -e 's/src\/ggml-kompute\.cpp/ggml/src/ggml-kompute.cpp/g' \
+        -e 's/src\/ggml-metal\.m/ggml/src/ggml-metal.m/g' \
+        -e 's/src\/ggml-quants\.c/ggml/src/ggml-quants.c/g' \
+        -e 's/src\/ggml-quants\.h/ggml/src/ggml-quants.h/g' \
+        -e 's/src\/ggml-rpc\.cpp/ggml/src/ggml-rpc.cpp/g' \
+        -e 's/src\/ggml-sycl\.cpp/ggml/src/ggml-sycl.cpp/g' \
+        -e 's/src\/ggml-vulkan\.cpp/ggml/src/ggml-vulkan.cpp/g' \
+        -e 's/include\/ggml\.h/ggml/include/ggml.h/g' \
+        -e 's/include\/ggml-alloc\.h/ggml/include/ggml-alloc.h/g' \
+        -e 's/include\/ggml-backend\.h/ggml/include/ggml-backend.h/g' \
+        -e 's/include\/ggml-blas\.h/ggml/include/ggml-blas.h/g' \
+        -e 's/include\/ggml-cuda\.h/ggml/include/ggml-cuda.h/g' \
+        -e 's/include\/ggml-kompute\.h/ggml/include/ggml-kompute.h/g' \
+        -e 's/include\/ggml-metal\.h/ggml/include/ggml-metal.h/g' \
+        -e 's/include\/ggml-rpc\.h/ggml/include/ggml-rpc.h/g' \
+        -e 's/include\/ggml-sycl\.h/ggml/include/ggml-sycl.h/g' \
+        -e 's/include\/ggml-vulkan\.h/ggml/include/ggml-vulkan.h/g' \
+        -e 's/tests\/test-opt\.cpp/tests\/test-opt.cpp/g' \
+        -e 's/tests\/test-grad0\.cpp/tests\/test-grad0.cpp/g' \
+        -e 's/tests\/test-quantize-fns\.cpp/tests\/test-quantize-fns.cpp/g' \
+        -e 's/tests\/test-quantize-perf\.cpp/tests\/test-quantize-perf.cpp/g' \
+        -e 's/tests\/test-backend-ops\.cpp/tests\/test-backend-ops.cpp/g' \
+        -e 's/LICENSE/LICENSE/g' \
+        -e 's/scripts\/gen-authors\.sh/scripts\/gen-authors.sh/g' \
        > ggml-src.patch.tmp
    mv ggml-src.patch.tmp ggml-src.patch

@@ -1 +1 @@
-5378ea0d3c2f25bcd330ecb226ad2db454be86d0
+5653a195935ea3ac54652644c9daf154dbc1571b
@@ -217,7 +217,6 @@ enum llm_arch {
    LLM_ARCH_INTERNLM2,
    LLM_ARCH_MINICPM,
    LLM_ARCH_GEMMA,
-    LLM_ARCH_GEMMA2,
    LLM_ARCH_STARCODER2,
    LLM_ARCH_MAMBA,
    LLM_ARCH_XVERSE,
@@ -228,7 +227,6 @@ enum llm_arch {
    LLM_ARCH_DEEPSEEK2,
    LLM_ARCH_BITNET,
    LLM_ARCH_T5,
-    LLM_ARCH_JAIS,
    LLM_ARCH_UNKNOWN,
 };

@@ -259,7 +257,6 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
    { LLM_ARCH_INTERNLM2,       "internlm2"    },
    { LLM_ARCH_MINICPM,         "minicpm"      },
    { LLM_ARCH_GEMMA,           "gemma"        },
-    { LLM_ARCH_GEMMA2,          "gemma2"       },
    { LLM_ARCH_STARCODER2,      "starcoder2"   },
    { LLM_ARCH_MAMBA,           "mamba"        },
    { LLM_ARCH_XVERSE,          "xverse"       },
@@ -270,7 +267,6 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
    { LLM_ARCH_DEEPSEEK2,       "deepseek2"    },
    { LLM_ARCH_BITNET,          "bitnet"       },
    { LLM_ARCH_T5,              "t5"           },
-    { LLM_ARCH_JAIS,            "jais"         },
    { LLM_ARCH_UNKNOWN,         "(unknown)"    },
 };

@@ -304,8 +300,6 @@ enum llm_kv {
    LLM_KV_POOLING_TYPE,
    LLM_KV_LOGIT_SCALE,
    LLM_KV_DECODER_START_TOKEN_ID,
-    LLM_KV_ATTN_LOGIT_SOFTCAPPING,
-    LLM_KV_FINAL_LOGIT_SOFTCAPPING,

    LLM_KV_ATTENTION_HEAD_COUNT,
    LLM_KV_ATTENTION_HEAD_COUNT_KV,
@@ -319,7 +313,6 @@ enum llm_kv {
    LLM_KV_ATTENTION_Q_LORA_RANK,
    LLM_KV_ATTENTION_KV_LORA_RANK,
    LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT,
-    LLM_KV_ATTENTION_SLIDING_WINDOW,

    LLM_KV_ROPE_DIMENSION_COUNT,
    LLM_KV_ROPE_FREQ_BASE,
@@ -397,8 +390,6 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
    { LLM_KV_POOLING_TYPE ,                     "%s.pooling_type"                      },
    { LLM_KV_LOGIT_SCALE,                       "%s.logit_scale"                       },
    { LLM_KV_DECODER_START_TOKEN_ID,            "%s.decoder_start_token_id"            },
-    { LLM_KV_ATTN_LOGIT_SOFTCAPPING,            "%s.attn_logit_softcapping"            },
-    { LLM_KV_FINAL_LOGIT_SOFTCAPPING,           "%s.final_logit_softcapping"           },

    { LLM_KV_ATTENTION_HEAD_COUNT,             "%s.attention.head_count"             },
    { LLM_KV_ATTENTION_HEAD_COUNT_KV,          "%s.attention.head_count_kv"          },
@@ -412,7 +403,6 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
    { LLM_KV_ATTENTION_Q_LORA_RANK,            "%s.attention.q_lora_rank"            },
    { LLM_KV_ATTENTION_KV_LORA_RANK,           "%s.attention.kv_lora_rank"           },
    { LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT, "%s.attention.relative_buckets_count" },
-    { LLM_KV_ATTENTION_SLIDING_WINDOW,         "%s.attention.sliding_window"         },

    { LLM_KV_ROPE_DIMENSION_COUNT,          "%s.rope.dimension_count"                 },
    { LLM_KV_ROPE_FREQ_BASE,                "%s.rope.freq_base"                       },
@@ -488,12 +478,10 @@ enum llm_tensor {
    LLM_TENSOR_ATTN_NORM,
    LLM_TENSOR_ATTN_NORM_2,
    LLM_TENSOR_ATTN_OUT_NORM,
-    LLM_TENSOR_ATTN_POST_NORM,
    LLM_TENSOR_ATTN_ROT_EMBD,
    LLM_TENSOR_FFN_GATE_INP,
    LLM_TENSOR_FFN_GATE_INP_SHEXP,
    LLM_TENSOR_FFN_NORM,
-    LLM_TENSOR_FFN_POST_NORM,
    LLM_TENSOR_FFN_GATE,
    LLM_TENSOR_FFN_DOWN,
    LLM_TENSOR_FFN_UP,
@@ -1016,24 +1004,6 @@ static const std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NA
            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
        },
    },
-    {
-        LLM_ARCH_GEMMA2,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
-            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
-            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
-        },
-    },
    {
        LLM_ARCH_STARCODER2,
        {
@@ -1238,21 +1208,6 @@ static const std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NA
            { LLM_TENSOR_ENC_FFN_UP,           "enc.blk.%d.ffn_up" },
        },
    },
-    {
-        LLM_ARCH_JAIS,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
-            { LLM_TENSOR_OUTPUT,          "output" },
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
-            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
-        },
-    },
    {
        LLM_ARCH_UNKNOWN,
        {
@@ -2052,7 +2007,6 @@ enum e_model {
    MODEL_410M,
    MODEL_0_5B,
    MODEL_1B,
-    MODEL_1_3B,
    MODEL_1_4B,
    MODEL_2B,
    MODEL_2_8B,
@@ -2084,9 +2038,6 @@ enum e_model {
    MODEL_8x22B,
    MODEL_16x12B,
    MODEL_10B_128x3_66B,
-    MODEL_57B_A14B,
-    MODEL_9B,
-    MODEL_27B,
 };

 static const size_t kiB = 1024;
@@ -2105,7 +2056,6 @@ struct llama_hparams {
    uint32_t n_head_kv;
    uint32_t n_layer;
    uint32_t n_rot;
-    uint32_t n_swa = 0; // sliding window attention (SWA)
    uint32_t n_embd_head_k; // dimension of keys (d_k). d_q is assumed to be the same, but there are n_head q heads, and only n_head_kv k-v heads
    uint32_t n_embd_head_v; // dimension of values (d_v) aka n_embd_head
    uint32_t n_ff;
@@ -2124,9 +2074,6 @@ struct llama_hparams {
    float f_norm_eps;
    float f_norm_rms_eps;

-    float f_attn_logit_softcapping = 50.0f;
-    float f_final_logit_softcapping = 30.0f;
-
    float    rope_attn_factor = 1.0f;
    float    rope_freq_base_train;
    float    rope_freq_scale_train;
@@ -2143,9 +2090,8 @@ struct llama_hparams {
    float f_max_alibi_bias = 0.0f;
    float f_logit_scale    = 0.0f;

-    bool causal_attn   = true;
-    bool use_alibi     = false;
-    bool attn_soft_cap = false;
+    bool causal_attn = true;
+    bool use_alibi   = false;

    enum llama_pooling_type      pooling_type            = LLAMA_POOLING_TYPE_NONE;
    enum llama_rope_type         rope_type               = LLAMA_ROPE_TYPE_NONE;
@@ -2160,7 +2106,6 @@ struct llama_hparams {
        if (this->n_head_kv     != other.n_head_kv)     return true;
        if (this->n_layer       != other.n_layer)       return true;
        if (this->n_rot         != other.n_rot)         return true;
-        if (this->n_swa         != other.n_swa)         return true;
        if (this->n_embd_head_k != other.n_embd_head_k) return true;
        if (this->n_embd_head_v != other.n_embd_head_v) return true;
        if (this->n_ff          != other.n_ff)          return true;
@@ -2269,7 +2214,6 @@ struct llama_layer {
    struct ggml_tensor * attn_q_a_norm;
    struct ggml_tensor * attn_kv_a_norm;
    struct ggml_tensor * attn_sub_norm;
-    struct ggml_tensor * attn_post_norm;
    struct ggml_tensor * ffn_sub_norm;

    // attention
@@ -2293,7 +2237,6 @@ struct llama_layer {
    // normalization
    struct ggml_tensor * ffn_norm;
    struct ggml_tensor * ffn_norm_b;
-    struct ggml_tensor * ffn_post_norm;
    struct ggml_tensor * layer_out_norm;
    struct ggml_tensor * layer_out_norm_b;
    struct ggml_tensor * ffn_norm_exps;
@@ -2671,18 +2614,17 @@ struct llama_context {
    void *              abort_callback_data = nullptr;

    // input tensors
-    struct ggml_tensor * inp_tokens;      // I32 [n_batch]
-    struct ggml_tensor * inp_embd;        // F32 [n_embd, n_batch]
-    struct ggml_tensor * inp_pos;         // I32 [n_batch]
-    struct ggml_tensor * inp_out_ids;     // I32 [n_outputs]
-    struct ggml_tensor * inp_KQ_mask;     // F32 [kv_size, n_batch]
-    struct ggml_tensor * inp_KQ_mask_swa; // F32 [kv_size, n_batch]
-    struct ggml_tensor * inp_K_shift;     // I32 [kv_size]
-    struct ggml_tensor * inp_mean;        // F32 [n_batch, n_batch]
-    struct ggml_tensor * inp_cls;         // I32 [n_batch]
-    struct ggml_tensor * inp_s_copy;      // I32 [kv_size]
-    struct ggml_tensor * inp_s_mask;      // F32 [1, n_kv]
-    struct ggml_tensor * inp_s_seq;       // I32 [n_kv, n_batch]
+    struct ggml_tensor * inp_tokens;    // I32 [n_batch]
+    struct ggml_tensor * inp_embd;      // F32 [n_embd, n_batch]
+    struct ggml_tensor * inp_pos;       // I32 [n_batch]
+    struct ggml_tensor * inp_out_ids;   // I32 [n_outputs]
+    struct ggml_tensor * inp_KQ_mask;   // F32 [kv_size, n_batch]
+    struct ggml_tensor * inp_K_shift;   // I32 [kv_size]
+    struct ggml_tensor * inp_mean;      // F32 [n_batch, n_batch]
+    struct ggml_tensor * inp_cls;       // I32 [n_batch]
+    struct ggml_tensor * inp_s_copy;    // I32 [kv_size]
+    struct ggml_tensor * inp_s_mask;    // F32 [1, n_kv]
+    struct ggml_tensor * inp_s_seq;     // I32 [n_kv, n_batch]

    // control vectors
    struct llama_control_vector cvec;
@@ -4294,7 +4236,6 @@ static const char * llama_model_type_name(e_model type) {
        case MODEL_410M:          return "410M";
        case MODEL_0_5B:          return "0.5B";
        case MODEL_1B:            return "1B";
-        case MODEL_1_3B:          return "1.3B";
        case MODEL_1_4B:          return "1.4B";
        case MODEL_2B:            return "2B";
        case MODEL_2_8B:          return "2.8B";
@@ -4326,9 +4267,6 @@ static const char * llama_model_type_name(e_model type) {
        case MODEL_8x22B:         return "8x22B";
        case MODEL_16x12B:        return "16x12B";
        case MODEL_10B_128x3_66B: return "10B+128x3.66B";
-        case MODEL_57B_A14B:      return "57B.A14B";
-        case MODEL_9B:            return "9B";
-        case MODEL_27B:           return "27B";
        default:                  return "?B";
    }
 }
@@ -4650,7 +4588,6 @@ static void llm_load_hparams(
                ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
                switch (hparams.n_layer) {
                    case 24: model.type = e_model::MODEL_A2_7B; break;
-                    case 28: model.type = e_model::MODEL_57B_A14B; break;
                    default: model.type = e_model::MODEL_UNKNOWN;
                }
            } break;
@@ -4731,21 +4668,6 @@ static void llm_load_hparams(
                    default: model.type = e_model::MODEL_UNKNOWN;
               }
            } break;
-        case LLM_ARCH_GEMMA2:
-            {
-                hparams.n_swa = 4096; // default value of gemma 2
-                ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa, false);
-                ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
-                ml.get_key(LLM_KV_ATTN_LOGIT_SOFTCAPPING, hparams.f_attn_logit_softcapping, false);
-                ml.get_key(LLM_KV_FINAL_LOGIT_SOFTCAPPING, hparams.f_final_logit_softcapping, false);
-                hparams.attn_soft_cap = true;
-
-                switch (hparams.n_layer) {
-                    case 42: model.type = e_model::MODEL_9B; break;
-                    case 46: model.type = e_model::MODEL_27B; break;
-                    default: model.type = e_model::MODEL_UNKNOWN;
-               }
-            } break;
        case LLM_ARCH_STARCODER2:
            {
                ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
@@ -4917,18 +4839,6 @@ static void llm_load_hparams(
                    default: model.type = e_model::MODEL_UNKNOWN;
                }
            } break;
-        case LLM_ARCH_JAIS:
-            {
-                ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
-                ml.get_key(LLM_KV_ATTENTION_MAX_ALIBI_BIAS, hparams.f_max_alibi_bias);
-
-                switch (hparams.n_layer) {
-                    case 24: model.type = e_model::MODEL_1_3B; break;
-                    case 40: model.type = e_model::MODEL_13B; break;
-                    /* TODO: add variants */
-                    default: model.type = e_model::MODEL_UNKNOWN;
-                }
-            } break;
        default: (void)0;
    }

@@ -5157,12 +5067,6 @@ static void llm_load_vocab(
            } else if (
                tokenizer_pre == "poro-chat") {
                vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_PORO;
-            } else if (
-                tokenizer_pre == "viking") {
-                vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_VIKING;
-            } else if (
-                tokenizer_pre == "jais") {
-                vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_JAIS;
            } else {
                throw std::runtime_error(format("unknown pre-tokenizer type: '%s'", tokenizer_pre.c_str()));
            }
@@ -5248,10 +5152,10 @@ static void llm_load_vocab(
        if (gen_name.find("code") != std::string::npos) {
            if (model.arch == LLM_ARCH_LLAMA
              && 32010 < vocab.id_to_token.size()
-              && vocab.id_to_token[32007].text.find("<PRE>") != std::string::npos
-              && vocab.id_to_token[32008].text.find("<SUF>") != std::string::npos
-              && vocab.id_to_token[32009].text.find("<MID>") != std::string::npos
-              && vocab.id_to_token[32010].text.find("<EOT>") != std::string::npos) {
+              && vocab.id_to_token[32007].text == "<PRE>"
+              && vocab.id_to_token[32008].text == "<SUF>"
+              && vocab.id_to_token[32009].text == "<MID>"
+              && vocab.id_to_token[32010].text == "<EOT>") {
                vocab.special_prefix_id = 32007;
                vocab.special_suffix_id = 32008;
                vocab.special_middle_id = 32009;
@@ -5460,7 +5364,6 @@ static void llm_load_print_meta(llama_model_loader & ml, llama_model & model) {
    LLAMA_LOG_INFO("%s: n_head_kv        = %u\n",     __func__, hparams.n_head_kv);
    LLAMA_LOG_INFO("%s: n_layer          = %u\n",     __func__, hparams.n_layer);
    LLAMA_LOG_INFO("%s: n_rot            = %u\n",     __func__, hparams.n_rot);
-    LLAMA_LOG_INFO("%s: n_swa            = %u\n",     __func__, hparams.n_swa);
    LLAMA_LOG_INFO("%s: n_embd_head_k    = %u\n",     __func__, hparams.n_embd_head_k);
    LLAMA_LOG_INFO("%s: n_embd_head_v    = %u\n",     __func__, hparams.n_embd_head_v);
    LLAMA_LOG_INFO("%s: n_gqa            = %u\n",     __func__, hparams.n_gqa());
@@ -6603,40 +6506,6 @@ static bool llm_load_tensors(
                        layer.ffn_down = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd});
                    }
                } break;
-            case LLM_ARCH_GEMMA2:
-                {
-                    model.tok_embd = ml.create_tensor(ctx_input, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
-
-                    // output
-                    model.output_norm = ml.create_tensor(ctx_output, tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd});
-                    model.output      = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD,  "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); // same as tok_embd, duplicated to allow offloading
-
-                    const int64_t n_ff          = hparams.n_ff;
-                    const int64_t n_embd_head_k = hparams.n_embd_head_k;
-                    const int64_t n_embd_k_gqa  = hparams.n_embd_k_gqa();
-                    const int64_t n_embd_v_gqa  = hparams.n_embd_v_gqa();
-
-                    for (uint32_t i = 0; i < n_layer; ++i) {
-                        ggml_context * ctx_layer = ctx_for_layer(i);
-                        ggml_context * ctx_split = ctx_for_layer_split(i);
-
-                        auto & layer = model.layers[i];
-
-                        layer.attn_norm = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd});
-
-                        layer.wq = ml.create_tensor(ctx_split, tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd_head_k * hparams.n_head});
-                        layer.wk = ml.create_tensor(ctx_split, tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_k_gqa});
-                        layer.wv = ml.create_tensor(ctx_split, tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_v_gqa});
-                        layer.wo = ml.create_tensor(ctx_split, tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * hparams.n_head, n_embd});
-                        layer.attn_post_norm = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_POST_NORM, "weight", i), {n_embd});
-
-                        layer.ffn_norm = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd});
-                        layer.ffn_gate = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff});
-                        layer.ffn_up   = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff});
-                        layer.ffn_down = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd});
-                        layer.ffn_post_norm = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_POST_NORM, "weight", i), {n_embd});
-                    }
-                } break;
            case LLM_ARCH_STARCODER2:
                {
                    model.tok_embd = ml.create_tensor(ctx_input, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
@@ -6996,44 +6865,6 @@ static bool llm_load_tensors(
                        layer.ffn_up_scale = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_UP, "scale", i), {1});
                    }
                } break;
-            case LLM_ARCH_JAIS:
-                {
-                    model.tok_embd = ml.create_tensor(ctx_input, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
-
-                    // Output
-                    {
-                        model.output_norm   = ml.create_tensor(ctx_output,       tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd});
-                        model.output_norm_b = ml.create_tensor(ctx_output,       tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd});
-                        model.output        = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab});
-                    }
-                    for (int i = 0; i < n_layer; ++i) {
-                        ggml_context * ctx_layer = ctx_for_layer(i);
-                        ggml_context * ctx_split = ctx_for_layer_split(i);
-
-                        auto & layer = model.layers[i];
-
-                        layer.attn_norm   = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_NORM,   "weight", i), {n_embd});
-                        layer.attn_norm_b = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_NORM,   "bias", i),   {n_embd});
-
-                        layer.wqkv = ml.create_tensor(ctx_split, tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa});
-                        layer.bqkv = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_QKV, "bias", i),   {n_embd + 2*n_embd_gqa});
-
-                        layer.wo   = ml.create_tensor(ctx_split, tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd});
-                        layer.bo   = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_OUT, "bias", i),   {n_embd});
-
-                        layer.ffn_norm   = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd});
-                        layer.ffn_norm_b = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_NORM, "bias", i),   {n_embd});
-
-                        layer.ffn_down   = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd});
-                        layer.ffn_down_b = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_DOWN, "bias", i),   {n_embd});
-
-                        layer.ffn_gate     = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_GATE,   "weight", i), {n_embd, n_ff});
-                        layer.ffn_gate_b   = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_GATE,   "bias", i),   {n_ff});
-
-                        layer.ffn_up     = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff});
-                        layer.ffn_up_b   = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_UP,   "bias", i),   {n_ff});
-                    }
-                } break;
            default:
                throw std::runtime_error("unknown architecture");
        }
@@ -7670,12 +7501,6 @@ static struct ggml_tensor * llm_build_kqv(
            kq = ggml_scale(ctx, kq, 30);
        }

-        if (hparams.attn_soft_cap) {
-            kq = ggml_scale(ctx, kq, 1.0f / hparams.f_attn_logit_softcapping);
-            kq = ggml_tanh(ctx, kq);
-            kq = ggml_scale(ctx, kq, hparams.f_attn_logit_softcapping);
-        }
-
        kq = ggml_soft_max_ext(ctx, kq, kq_mask, kq_scale, hparams.f_max_alibi_bias);
        cb(kq, "kq_soft_max_ext", il);

@@ -7855,18 +7680,17 @@ struct llm_build_context {

        ctx0 = ggml_init(params);

-        lctx.inp_tokens      = nullptr;
-        lctx.inp_embd        = nullptr;
-        lctx.inp_pos         = nullptr;
-        lctx.inp_out_ids     = nullptr;
-        lctx.inp_KQ_mask     = nullptr;
-        lctx.inp_KQ_mask_swa = nullptr;
-        lctx.inp_K_shift     = nullptr;
-        lctx.inp_mean        = nullptr;
-        lctx.inp_cls         = nullptr;
-        lctx.inp_s_copy      = nullptr;
-        lctx.inp_s_mask      = nullptr;
-        lctx.inp_s_seq       = nullptr;
+        lctx.inp_tokens  = nullptr;
+        lctx.inp_embd    = nullptr;
+        lctx.inp_pos     = nullptr;
+        lctx.inp_out_ids = nullptr;
+        lctx.inp_KQ_mask = nullptr;
+        lctx.inp_K_shift = nullptr;
+        lctx.inp_mean    = nullptr;
+        lctx.inp_cls     = nullptr;
+        lctx.inp_s_copy  = nullptr;
+        lctx.inp_s_mask  = nullptr;
+        lctx.inp_s_seq   = nullptr;
    }

    void free() {
@@ -7885,6 +7709,7 @@ struct llm_build_context {
        cb(lctx.inp_K_shift, "K_shift", -1);
        ggml_set_input(lctx.inp_K_shift);

+
        for (int il = 0; il < n_layer; ++il) {
            struct ggml_tensor * rope_factors = build_rope_factors(il);
            struct ggml_tensor * tmp =
@@ -8019,27 +7844,16 @@ struct llm_build_context {
    }

    struct ggml_tensor * build_inp_KQ_mask(bool causal = true) {
-        lctx.inp_KQ_mask = causal
-            ? ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv,     GGML_PAD(n_tokens, GGML_KQ_MASK_PAD))
-            : ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
+        if (causal) {
+            lctx.inp_KQ_mask = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv,     GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
+        } else {
+            lctx.inp_KQ_mask = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
+        }
        cb(lctx.inp_KQ_mask, "KQ_mask", -1);
        ggml_set_input(lctx.inp_KQ_mask);
-
        return flash_attn ? ggml_cast(ctx0, lctx.inp_KQ_mask, GGML_TYPE_F16) : lctx.inp_KQ_mask;
    }

-    struct ggml_tensor * build_inp_KQ_mask_swa(bool causal = true) {
-        GGML_ASSERT(hparams.n_swa > 0);
-
-        lctx.inp_KQ_mask_swa = causal
-            ? ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv,     GGML_PAD(n_tokens, GGML_KQ_MASK_PAD))
-            : ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
-        cb(lctx.inp_KQ_mask_swa, "KQ_mask_swa", -1);
-        ggml_set_input(lctx.inp_KQ_mask_swa);
-
-        return flash_attn ? ggml_cast(ctx0, lctx.inp_KQ_mask_swa, GGML_TYPE_F16) : lctx.inp_KQ_mask_swa;
-    }
-
    struct ggml_tensor * build_inp_mean() {
        lctx.inp_mean = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, n_tokens);
        cb(lctx.inp_mean, "inp_mean", -1);
@@ -11103,136 +10917,6 @@ struct llm_build_context {
        return gf;
    }

-    struct ggml_cgraph * build_gemma2() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false);
-
-        const int64_t n_embd_head_k = hparams.n_embd_head_k;
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, batch, model.tok_embd, cb);
-
-        inpL = ggml_scale(ctx0, inpL, sqrtf(n_embd));
-        cb(inpL, "inp_scaled", -1);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        // gemma 2 requires different mask for layers using sliding window (SWA)
-        struct ggml_tensor * KQ_mask     = build_inp_KQ_mask(true);
-        struct ggml_tensor * KQ_mask_swa = build_inp_KQ_mask_swa(true);
-
-        for (int il = 0; il < n_layer; ++il) {
-            // (il % 2) layers use SWA
-            struct ggml_tensor * KQ_mask_l = (il % 2 == 0) ? KQ_mask_swa : KQ_mask;
-
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = ggml_mul_mat(ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-
-                struct ggml_tensor * Kcur = ggml_mul_mat(ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-
-                struct ggml_tensor * Vcur = ggml_mul_mat(ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head_k, n_head,    n_tokens), inp_pos, nullptr,
-                        n_embd_head_k, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
-                cb(Qcur, "Qcur", il);
-
-                Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd / n_head)));
-                cb(Qcur, "Qcur_scaled", il);
-
-                Kcur = ggml_rope_ext(
-                        ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head_k, n_head_kv, n_tokens), inp_pos, nullptr,
-                        n_embd_head_k, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
-                cb(Kcur, "Kcur", il);
-
-                cur = llm_build_kv(ctx0, model, hparams, cparams, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask_l, n_tokens, kv_head, n_kv, 1.0f, cb, il);
-            }
-
-            cur = llm_build_norm(ctx0, cur, hparams,
-                    model.layers[il].attn_post_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_post_norm", il);
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-
-            struct ggml_tensor * sa_out = ggml_add(ctx0, cur, inpL);
-            cb(sa_out, "sa_out", il);
-
-            cur = llm_build_norm(ctx0, sa_out, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
-
-            // feed-forward network
-            {
-                cur = llm_build_ffn(ctx0, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_PAR, cb, il);
-                cb(cur, "ffn_out", il);
-            }
-
-            cur = llm_build_norm(ctx0, cur, hparams,
-                model.layers[il].ffn_post_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-            cb(cur, "ffn_post_norm", -1);
-
-            cur = ggml_add(ctx0, cur, sa_out);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = ggml_mul_mat(ctx0, model.output, cur);
-
-        // final logit soft-capping
-        cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_final_logit_softcapping);
-        cur = ggml_tanh(ctx0, cur);
-        cur = ggml_scale(ctx0, cur, hparams.f_final_logit_softcapping);
-
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-
    struct ggml_cgraph * build_starcoder2() {
        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false);

@@ -12426,97 +12110,6 @@ struct llm_build_context {
        return gf;
    }

-    struct ggml_cgraph * build_jais() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false);
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, batch, model.tok_embd, cb);
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                cur = ggml_mul_mat(ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
-
-                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*cur->nb[0]*(n_embd)));
-                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*cur->nb[0]*(n_embd)));
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*cur->nb[0]*(n_embd + n_embd_gqa)));
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-
-                cur = llm_build_kv(ctx0, model, hparams, cparams, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/float(n_embd_head), cb, il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-
-            // add the input
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // FF
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = llm_build_ffn(ctx0, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-                cb(cur, "ffn_out", il);
-            }
-
-            inpL = ggml_add(ctx0, cur, ffn_inp);
-            cb(inpL, "l_out", il);
-        }
-
-        cur = llm_build_norm(ctx0, inpL, hparams,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        cur = ggml_mul_mat(ctx0, model.output, cur);
-
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
 };

 static struct ggml_cgraph * llama_build_graph_defrag(llama_context & lctx, const std::vector<uint32_t> & ids) {
@@ -12704,10 +12297,6 @@ static struct ggml_cgraph * llama_build_graph(
            {
                result = llm.build_gemma();
            } break;
-        case LLM_ARCH_GEMMA2:
-            {
-                result = llm.build_gemma2();
-            } break;
        case LLM_ARCH_STARCODER2:
            {
                result = llm.build_starcoder2();
@@ -12748,10 +12337,6 @@ static struct ggml_cgraph * llama_build_graph(
            {
                result = llm.build_bitnet();
            } break;
-        case LLM_ARCH_JAIS:
-            {
-                result = llm.build_jais();
-            } break;
        default:
            GGML_ASSERT(false);
    }
@@ -12861,12 +12446,7 @@ static void llama_set_inputs(llama_context & lctx, const llama_batch & batch) {

            GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_KQ_mask->buffer));

-            float * data     = (float *) lctx.inp_KQ_mask->data;
-            float * data_swa = nullptr;
-
-            if (lctx.inp_KQ_mask_swa) {
-                data_swa = (float *) lctx.inp_KQ_mask_swa->data;
-            }
+            float * data = (float *) lctx.inp_KQ_mask->data;

            // For causal attention, use only the previous KV cells
            // of the correct sequence for each token of the batch.
@@ -12888,14 +12468,6 @@ static void llama_set_inputs(llama_context & lctx, const llama_batch & batch) {
                            }
                        }
                        data[h*(n_kv*n_tokens) + j*n_kv + i] = f;
-
-                        // may need to cut off old tokens for sliding window
-                        if (data_swa) {
-                            if (pos - lctx.kv_self.cells[i].pos >= (int32_t)hparams.n_swa) {
-                                f = -INFINITY;
-                            }
-                            data_swa[h*(n_kv*n_tokens) + j*n_kv + i] = f;
-                        }
                    }
                }

@@ -14114,7 +13686,6 @@ struct llm_tokenizer_bpe {
                break;
            case LLAMA_VOCAB_PRE_TYPE_GPT2:
            case LLAMA_VOCAB_PRE_TYPE_OLMO:
-            case LLAMA_VOCAB_PRE_TYPE_JAIS:
                regex_exprs = {
                    "'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)",
                };
@@ -14132,12 +13703,6 @@ struct llm_tokenizer_bpe {
                    " ?[^(\\s|.,!?…。，、।۔،)]+",
                };
                break;
-            case LLAMA_VOCAB_PRE_TYPE_VIKING:
-                regex_exprs = {
-                    "\\p{N}",
-                    " ?[^(\\s|.,!?…。，、।۔،)]+",
-                };
-                break;
            default:
                // default regex for BPE tokenization pre-processing
                regex_exprs = {
@@ -17607,12 +17172,6 @@ struct llama_context * llama_new_context_with_model(
        params.flash_attn = false;
    }

-    if (params.flash_attn && model->hparams.attn_soft_cap) {
-        LLAMA_LOG_WARN("%s: flash_attn is not compatible with attn_soft_cap - forcing off\n", __func__);
-        params.flash_attn = false;
-    }
-
-
    if (params.flash_attn && model->hparams.n_embd_head_k != model->hparams.n_embd_head_v) {
        LLAMA_LOG_WARN("%s: flash_attn requires n_embd_head_k == n_embd_head_v - forcing off\n", __func__);
        params.flash_attn = false;
@@ -17994,7 +17553,6 @@ enum llama_rope_type llama_rope_type(const struct llama_model * model) {
        case LLM_ARCH_MAMBA:
        case LLM_ARCH_JINA_BERT_V2:
        case LLM_ARCH_T5:
-        case LLM_ARCH_JAIS:
            return LLAMA_ROPE_TYPE_NONE;

        // use what we call a normal RoPE, operating on pairs of consecutive head values
@@ -18027,7 +17585,6 @@ enum llama_rope_type llama_rope_type(const struct llama_model * model) {
        case LLM_ARCH_PHI2:
        case LLM_ARCH_PHI3:
        case LLM_ARCH_GEMMA:
-        case LLM_ARCH_GEMMA2:
        case LLM_ARCH_STARCODER2:
        case LLM_ARCH_GPTNEOX:
            return LLAMA_ROPE_TYPE_NEOX;
@@ -19848,10 +19405,7 @@ static int32_t llama_chat_apply_template_internal(
    std::string & dest, bool add_ass) {
    // Taken from the research: https://github.com/ggerganov/llama.cpp/issues/5527
    std::stringstream ss;
-    auto tmpl_contains = [&tmpl](std::string haystack) -> bool {
-        return tmpl.find(haystack) != std::string::npos;
-    };
-    if (tmpl == "chatml" || tmpl_contains("<|im_start|>")) {
+    if (tmpl == "chatml" || tmpl.find("<|im_start|>") != std::string::npos) {
        // chatml template
        for (auto message : chat) {
            ss << "<|im_start|>" << message->role << "\n" << message->content << "<|im_end|>\n";
@@ -19859,16 +19413,16 @@ static int32_t llama_chat_apply_template_internal(
        if (add_ass) {
            ss << "<|im_start|>assistant\n";
        }
-    } else if (tmpl == "llama2" || tmpl == "mistral" || tmpl_contains("[INST]")) {
+    } else if (tmpl == "llama2" || tmpl == "mistral" || tmpl.find("[INST]") != std::string::npos) {
        // llama2 template and its variants
        // [variant] support system message
-        bool support_system_message = tmpl_contains("<<SYS>>") || tmpl == "mistral";
+        bool support_system_message = tmpl.find("<<SYS>>") != std::string::npos || tmpl == "mistral";
        // [variant] space before + after response
-        bool space_around_response = tmpl_contains("' ' + eos_token");
+        bool space_around_response = tmpl.find("' ' + eos_token") != std::string::npos;
        // [variant] add BOS inside history
-        bool add_bos_inside_history = tmpl_contains("bos_token + '[INST]");
+        bool add_bos_inside_history = tmpl.find("bos_token + '[INST]") != std::string::npos;
        // [variant] trim spaces from the input message
-        bool strip_message = tmpl_contains("content.strip()");
+        bool strip_message = tmpl.find("content.strip()") != std::string::npos;
        // construct the prompt
        bool is_inside_turn = true; // skip BOS at the beginning
        ss << "[INST] ";
@@ -19894,7 +19448,7 @@ static int32_t llama_chat_apply_template_internal(
            }
        }
        // llama2 templates seem to not care about "add_generation_prompt"
-    } else if (tmpl == "phi3" || (tmpl_contains("<|assistant|>") && tmpl_contains("<|end|>"))) {
+    } else if (tmpl == "phi3" || (tmpl.find("<|assistant|>") != std::string::npos && tmpl.find("<|end|>") != std::string::npos)) {
        // Phi 3
        for (auto message : chat) {
            std::string role(message->role);
@@ -19903,7 +19457,7 @@ static int32_t llama_chat_apply_template_internal(
        if (add_ass) {
            ss << "<|assistant|>\n";
        }
-    } else if (tmpl == "zephyr" || tmpl_contains("<|user|>")) {
+    } else if (tmpl == "zephyr" || tmpl.find("<|user|>") != std::string::npos) {
        // zephyr template
        for (auto message : chat) {
            ss << "<|" << message->role << "|>" << "\n" << message->content << "<|endoftext|>\n";
@@ -19911,7 +19465,7 @@ static int32_t llama_chat_apply_template_internal(
        if (add_ass) {
            ss << "<|assistant|>\n";
        }
-    } else if (tmpl == "monarch" || tmpl_contains("bos_token + message['role']")) {
+    } else if (tmpl == "monarch" || tmpl.find("bos_token + message['role']") != std::string::npos) {
        // mlabonne/AlphaMonarch-7B template (the <s> is included inside history)
        for (auto message : chat) {
            std::string bos = (message == chat.front()) ? "" : "<s>"; // skip BOS for first message
@@ -19920,7 +19474,7 @@ static int32_t llama_chat_apply_template_internal(
        if (add_ass) {
            ss << "<s>assistant\n";
        }
-    } else if (tmpl == "gemma" || tmpl == "gemma2" || tmpl_contains("<start_of_turn>")) {
+    } else if (tmpl == "gemma" || tmpl.find("<start_of_turn>") != std::string::npos) {
        // google/gemma-7b-it
        std::string system_prompt = "";
        for (auto message : chat) {
@@ -19942,7 +19496,7 @@ static int32_t llama_chat_apply_template_internal(
        if (add_ass) {
            ss << "<start_of_turn>model\n";
        }
-    } else if (tmpl == "orion" || tmpl_contains("'\\n\\nAssistant: ' + eos_token")) {
+    } else if (tmpl == "orion" || tmpl.find("'\\n\\nAssistant: ' + eos_token") != std::string::npos) {
        // OrionStarAI/Orion-14B-Chat
        std::string system_prompt = "";
        for (auto message : chat) {
@@ -19962,7 +19516,7 @@ static int32_t llama_chat_apply_template_internal(
                ss << message->content << "</s>";
            }
        }
-    } else if (tmpl == "openchat" || tmpl_contains("GPT4 Correct ")) {
+    } else if (tmpl == "openchat" || tmpl.find("GPT4 Correct ") != std::string::npos) {
        // openchat/openchat-3.5-0106,
        for (auto message : chat) {
            std::string role(message->role);
@@ -19976,13 +19530,13 @@ static int32_t llama_chat_apply_template_internal(
        if (add_ass) {
            ss << "GPT4 Correct Assistant:";
        }
-    } else if (tmpl == "vicuna" || tmpl == "vicuna-orca" || (tmpl_contains("USER: ") && tmpl_contains("ASSISTANT: "))) {
+    } else if (tmpl == "vicuna" || tmpl == "vicuna-orca" || (tmpl.find("USER: ") != std::string::npos && tmpl.find("ASSISTANT: ") != std::string::npos)) {
        // eachadea/vicuna-13b-1.1 (and Orca variant)
        for (auto message : chat) {
            std::string role(message->role);
            if (role == "system") {
                // Orca-Vicuna variant uses a system prefix
-                if (tmpl == "vicuna-orca" || tmpl_contains("SYSTEM: ")) {
+                if (tmpl == "vicuna-orca" || tmpl.find("SYSTEM: ") != std::string::npos) {
                    ss << "SYSTEM: " << message->content << "\n";
                } else {
                    ss << message->content << "\n\n";
@@ -19996,7 +19550,7 @@ static int32_t llama_chat_apply_template_internal(
        if (add_ass) {
            ss << "ASSISTANT:";
        }
-    } else if (tmpl == "deepseek" || (tmpl_contains("### Instruction:") && tmpl_contains("<|EOT|>"))) {
+    } else if (tmpl == "deepseek" || (tmpl.find("### Instruction:") != std::string::npos && tmpl.find("<|EOT|>") != std::string::npos)) {
        // deepseek-ai/deepseek-coder-33b-instruct
        for (auto message : chat) {
            std::string role(message->role);
@@ -20011,7 +19565,7 @@ static int32_t llama_chat_apply_template_internal(
        if (add_ass) {
            ss << "### Response:\n";
        }
-    } else if (tmpl == "command-r" || (tmpl_contains("<|START_OF_TURN_TOKEN|>") && tmpl_contains("<|USER_TOKEN|>"))) {
+    } else if (tmpl == "command-r" || (tmpl.find("<|START_OF_TURN_TOKEN|>") != std::string::npos && tmpl.find("<|USER_TOKEN|>") != std::string::npos)) {
        // CohereForAI/c4ai-command-r-plus
        for (auto message : chat) {
            std::string role(message->role);
@@ -20026,7 +19580,7 @@ static int32_t llama_chat_apply_template_internal(
        if (add_ass) {
            ss << "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>";
        }
-    } else if (tmpl == "llama3" || (tmpl_contains("<|start_header_id|>") && tmpl_contains("<|end_header_id|>"))) {
+    } else if (tmpl == "llama3" || (tmpl.find("<|start_header_id|>") != std::string::npos && tmpl.find("<|end_header_id|>") != std::string::npos)) {
        // Llama 3
        for (auto message : chat) {
            std::string role(message->role);
@@ -20035,33 +19589,6 @@ static int32_t llama_chat_apply_template_internal(
        if (add_ass) {
            ss << "<|start_header_id|>assistant<|end_header_id|>\n\n";
        }
-    } else if (tmpl == "minicpm" || tmpl_contains(u8"<用户>")) {
-        // MiniCPM-3B-OpenHermes-2.5-v2-GGUF
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "user") {
-                ss << u8"<用户>";
-                ss << trim(message->content);
-                ss << "<AI>";
-            } else {
-                ss << trim(message->content);
-            }
-        }
-    } else if (tmpl == "deepseek2" || tmpl_contains("'Assistant: ' + message['content'] + eos_token")) {
-        // DeepSeek-V2
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "system") {
-                ss << message->content << "\n\n";
-            } else if (role == "user") {
-                ss << "User: " << message->content << "\n\n";
-            } else if (role == "assistant") {
-                ss << "Assistant: " << message->content << u8"<｜end▁of▁sentence｜>";
-            }
-        }
-        if (add_ass) {
-            ss << "Assistant:";
-        }
    } else {
        // template not supported
        return -1;
@@ -7030,3 +7030,4 @@ const std::vector<range_nfd> unicode_ranges_nfd = {  // start, last, nfd
 {0x02FA1C, 0x02FA1C, 0x009F3B},
 {0x02FA1D, 0x02FA1D, 0x02A600},
 };
+
@@ -2052,7 +2052,6 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
        GGML_TYPE_IQ2_XS, GGML_TYPE_IQ2_S,
        GGML_TYPE_IQ3_XXS, GGML_TYPE_IQ1_S, GGML_TYPE_IQ1_M,
        GGML_TYPE_IQ4_NL, GGML_TYPE_IQ3_S, GGML_TYPE_IQ4_XS,
-        GGML_TYPE_BF16,
    };

    // unary ops
@@ -57,11 +57,7 @@ int main(void) {
        //Phi-3-medium
        "{% for message in messages %}{% if (message['role'] == 'user') %}{{'<|user|>' + '\n' + message['content'] + '<|end|>' + '\n' + '<|assistant|>' + '\n'}}{% elif (message['role'] == 'assistant') %}{{message['content'] + '<|end|>' + '\n'}}{% endif %}{% endfor %}",
        //Phi-3-vision
-        "{% for message in messages %}{{'<|' + message['role'] + '|>' + '\n' + message['content'] + '<|end|>\n' }}{% endfor %}{% if add_generation_prompt and messages[-1]['role'] != 'assistant' %}{{- '<|assistant|>\n' -}}{% endif %}",
-        // MiniCPM-3B-OpenHermes-2.5-v2-GGUF
-        u8"{% for message in messages %}{% if message['role'] == 'user' %}{{'<用户>' + message['content'].strip() + '<AI>'}}{% else %}{{message['content'].strip()}}{% endif %}{% endfor %}",
-        // DeepSeek-V2
-        "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{{ bos_token }}{% for message in messages %}{% if message['role'] == 'user' %}{{ 'User: ' + message['content'] + '\n\n' }}{% elif message['role'] == 'assistant' %}{{ 'Assistant: ' + message['content'] + eos_token }}{% elif message['role'] == 'system' %}{{ message['content'] + '\n\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ 'Assistant:' }}{% endif %}",
+        "{% for message in messages %}{{'<|' + message['role'] + '|>' + '\n' + message['content'] + '<|end|>\n' }}{% endfor %}{% if add_generation_prompt and messages[-1]['role'] != 'assistant' %}{{- '<|assistant|>\n' -}}{% endif %}"
    };
    std::vector<std::string> expected_output = {
        // teknium/OpenHermes-2.5-Mistral-7B
@@ -98,10 +94,6 @@ int main(void) {
        "<|system|>\nYou are a helpful assistant<|end|>\n<|user|>\nHello<|end|>\n<|assistant|>\nHi there<|end|>\n<|user|>\nWho are you<|end|>\n<|assistant|>\n   I am an assistant   <|end|>\n<|user|>\nAnother question<|end|>\n<|assistant|>\n",
        //Phi-3-vision
        "<|system|>\nYou are a helpful assistant<|end|>\n<|user|>\nHello<|end|>\n<|assistant|>\nHi there<|end|>\n<|user|>\nWho are you<|end|>\n<|assistant|>\n   I am an assistant   <|end|>\n<|user|>\nAnother question<|end|>\n<|assistant|>\n",
-        // MiniCPM-3B-OpenHermes-2.5-v2-GGUF
-        u8"You are a helpful assistant<用户>Hello<AI>Hi there<用户>Who are you<AI>I am an assistant<用户>Another question<AI>",
-        // DeepSeek-V2
-        u8"You are a helpful assistant\n\nUser: Hello\n\nAssistant: Hi there<｜end▁of▁sentence｜>User: Who are you\n\nAssistant:    I am an assistant   <｜end▁of▁sentence｜>User: Another question\n\nAssistant:",
    };
    std::vector<char> formatted_chat(1024);
    int32_t res;
@@ -142,9 +134,9 @@ int main(void) {
        std::cout << "fmt_single(" << tmpl << ")\n" << output << "\n-------------------------\n";
        return output;
    };
-    assert(fmt_single("chatml") == "\n<|im_start|>user\nHow are you<|im_end|>\n<|im_start|>assistant\n");
+    assert(fmt_single("chatml") == "<|im_start|>user\nHow are you<|im_end|>\n<|im_start|>assistant\n");
    assert(fmt_single("llama2") == "[INST] How are you [/INST]");
-    assert(fmt_single("gemma") == "\n<start_of_turn>user\nHow are you<end_of_turn>\n<start_of_turn>model\n");
+    assert(fmt_single("gemma") == "<start_of_turn>user\nHow are you<end_of_turn>\n<start_of_turn>model\n");
    assert(fmt_single("llama3") == "<|start_header_id|>user<|end_header_id|>\n\nHow are you<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n");

    return 0;
@@ -993,40 +993,6 @@ static void test_json_schema() {
        }
    );

-    test_schema(
-        "simple pattern",
-        // Schema
-        R"""({
-            "pattern": "^[a-zA-Z0-9_-]*$"
-        })""",
-        // Passing strings
-        {
-            R"""("")""",
-            R"""("He_llo-12")""",
-        },
-        // Failing strings
-        {
-            R"""("!")""",
-            R"""("Hello World")""",
-        }
-    );
-
-    test_schema(
-        "pattern with escapes",
-        // Schema
-        R"""({
-            "pattern": "^a\\^\\$\\.\\[\\]\\(\\)\\|\\{\\}\\*\\+\\?b$"
-        })""",
-        // Passing strings
-        {
-            R"""("a^$.[]()|{}*+?b")""",
-        },
-        // Failing strings
-        {
-            R"""("ab")""",
-        }
-    );
-
    test_schema(
        "",
        // Schema
@@ -1096,6 +1062,8 @@ static void test_json_schema() {
            R"""({ "number": 1600, "street_name": "Pennsylvania" })""",
            // "By extension, even an empty object is valid"
            R"""({})""",
+            // "By default, providing additional properties is valid"
+            R"""({ "number": 1600, "street_name": "Pennsylvania", "street_type":"Avenue", "direction":"NW"})""",
            R"""({ "number": 1600, "street_name": "Pennsylvania", "street_type": "Avenue" })""",
        },
        // Failing strings
@@ -1106,9 +1074,6 @@ static void test_json_schema() {
            R"""({ "street_name": "Pennsylvania", "number": 1600 })""",
            // Reorder properties
            R"""({ "number": "1600", "street_name": "Pennsylvania", "street_type":"Avenue"})""",
-            // "Additional properties default to false for generation, even though the spec says true.
-            R"""({ "number": 1600, "street_name": "Pennsylvania", "street_type":"Avenue", "direction":"NW"})""",
-
        }
    );

@@ -1120,15 +1120,28 @@ static void test_all(const std::string & lang, std::function<void(const TestCase
        R"""(
            alternative-0 ::= foo
            alternative-1 ::= bar
-            bar ::= "{" space  (bar-b-kv )? "}" space
+            array ::= "[" space ( value ("," space value)* )? "]" space
+            bar ::= "{" space  (bar-b-kv bar-b-rest | bar-additional-kv ( "," space bar-additional-kv )* )? "}" space
+            bar-additional-k ::= ["] ( [b] char+ | [^"b] char* )? ["] space
+            bar-additional-kv ::= bar-additional-k ":" space value
            bar-b-kv ::= "\"b\"" space ":" space number
+            bar-b-rest ::= ( "," space bar-additional-kv )*
+            boolean ::= ("true" | "false") space
+            char ::= [^"\\\x7F\x00-\x1F] | [\\] (["\\bfnrt] | "u" [0-9a-fA-F]{4})
            decimal-part ::= [0-9]{1,16}
-            foo ::= "{" space  (foo-a-kv )? "}" space
+            foo ::= "{" space  (foo-a-kv foo-a-rest | foo-additional-kv ( "," space foo-additional-kv )* )? "}" space
            foo-a-kv ::= "\"a\"" space ":" space number
+            foo-a-rest ::= ( "," space foo-additional-kv )*
+            foo-additional-k ::= ["] ( [a] char+ | [^"a] char* )? ["] space
+            foo-additional-kv ::= foo-additional-k ":" space value
            integral-part ::= [0] | [1-9] [0-9]{0,15}
+            null ::= "null" space
            number ::= ("-"? integral-part) ("." decimal-part)? ([eE] [-+]? integral-part)? space
+            object ::= "{" space ( string ":" space value ("," space string ":" space value)* )? "}" space
            root ::= alternative-0 | alternative-1
            space ::= | " " | "\n" [ \t]{0,20}
+            string ::= "\"" char* "\"" space
+            value ::= object | array | string | number | boolean | null
        )"""
    });

@@ -1164,15 +1177,25 @@ static void test_all(const std::string & lang, std::function<void(const TestCase
        })""",
        R"""(
            a-kv ::= "\"a\"" space ":" space number
+            additional-k ::= ["] ( [a] char+ | [b] char+ | [c] char+ | [d] char+ | [^"abcd] char* )? ["] space
+            additional-kv ::= additional-k ":" space value
+            array ::= "[" space ( value ("," space value)* )? "]" space
            b-kv ::= "\"b\"" space ":" space number
+            boolean ::= ("true" | "false") space
            c-kv ::= "\"c\"" space ":" space number
+            c-rest ::= ( "," space additional-kv )*
+            char ::= [^"\\\x7F\x00-\x1F] | [\\] (["\\bfnrt] | "u" [0-9a-fA-F]{4})
            d-kv ::= "\"d\"" space ":" space number
-            d-rest ::= ( "," space c-kv )?
+            d-rest ::= ( "," space c-kv )? c-rest
            decimal-part ::= [0-9]{1,16}
            integral-part ::= [0] | [1-9] [0-9]{0,15}
+            null ::= "null" space
            number ::= ("-"? integral-part) ("." decimal-part)? ([eE] [-+]? integral-part)? space
-            root ::= "{" space a-kv "," space b-kv ( "," space ( d-kv d-rest | c-kv ) )? "}" space
+            object ::= "{" space ( string ":" space value ("," space string ":" space value)* )? "}" space
+            root ::= "{" space a-kv "," space b-kv ( "," space ( d-kv d-rest | c-kv c-rest | additional-kv ( "," space additional-kv )* ) )? "}" space
            space ::= | " " | "\n" [ \t]{0,20}
+            string ::= "\"" char* "\"" space
+            value ::= object | array | string | number | boolean | null
        )"""
    });

@@ -1239,30 +1262,26 @@ int main() {
        }
    });

-    if (getenv("LLAMA_SKIP_TESTS_SLOW_ON_EMULATOR")) {
-        fprintf(stderr, "\033[33mWARNING: Skipping slow tests on emulator.\n\033[0m");
+    if (getenv("LLAMA_PYTHON_AVAILABLE") || (std::system("python -c \"import sys; exit(1) if sys.version_info < (3, 8) else print('Python version is sufficient')\"") == 0)) {
+        test_all("Python", [](const TestCase & tc) {
+            write("test-json-schema-input.tmp", tc.schema);
+            tc.verify_status(std::system(
+                "python ./examples/json_schema_to_grammar.py test-json-schema-input.tmp > test-grammar-output.tmp") == 0 ? SUCCESS : FAILURE);
+            tc.verify(read("test-grammar-output.tmp"));
+        });
    } else {
-        if (getenv("LLAMA_PYTHON_AVAILABLE") || (std::system("python -c \"import sys; exit(1) if sys.version_info < (3, 8) else print('Python version is sufficient')\"") == 0)) {
-            test_all("Python", [](const TestCase & tc) {
-                write("test-json-schema-input.tmp", tc.schema);
-                tc.verify_status(std::system(
-                    "python ./examples/json_schema_to_grammar.py test-json-schema-input.tmp > test-grammar-output.tmp") == 0 ? SUCCESS : FAILURE);
-                tc.verify(read("test-grammar-output.tmp"));
-            });
-        } else {
-            fprintf(stderr, "\033[33mWARNING: Python not found (min version required is 3.8), skipping Python JSON schema -> grammar tests.\n\033[0m");
-        }
+        fprintf(stderr, "\033[33mWARNING: Python not found (min version required is 3.8), skipping Python JSON schema -> grammar tests.\n\033[0m");
+    }

-        if (getenv("LLAMA_NODE_AVAILABLE") || (std::system("node --version") == 0)) {
-            test_all("JavaScript", [](const TestCase & tc) {
-                write("test-json-schema-input.tmp", tc.schema);
-                tc.verify_status(std::system(
-                    "node ./tests/run-json-schema-to-grammar.mjs test-json-schema-input.tmp > test-grammar-output.tmp") == 0 ? SUCCESS : FAILURE);
-                tc.verify(read("test-grammar-output.tmp"));
-            });
-        } else {
-            fprintf(stderr, "\033[33mWARNING: Node not found, skipping JavaScript JSON schema -> grammar tests.\n\033[0m");
-        }
+    if (getenv("LLAMA_NODE_AVAILABLE") || (std::system("node --version") == 0)) {
+        test_all("JavaScript", [](const TestCase & tc) {
+            write("test-json-schema-input.tmp", tc.schema);
+            tc.verify_status(std::system(
+                "node ./tests/run-json-schema-to-grammar.mjs test-json-schema-input.tmp > test-grammar-output.tmp") == 0 ? SUCCESS : FAILURE);
+            tc.verify(read("test-grammar-output.tmp"));
+        });
+    } else {
+        fprintf(stderr, "\033[33mWARNING: Node not found, skipping JavaScript JSON schema -> grammar tests.\n\033[0m");
    }

    test_all("Check Expectations Validity", [](const TestCase & tc) {
@@ -218,3 +218,4 @@ int main(int /*argc*/, const char ** /*argv*/) {

    return 0;
 }
+
				`@@ -7,3 +7,5 @@ set INPUT2="Building a website can be done in 10 simple steps:\nStep 1:"`


				`.\build\bin\main.exe -m models\llama-2-7b.Q4_0.gguf -p %INPUT2% -n 400 -e -ngl 33 -s 0`