wylab/llama.cpp
1
0
Fork 1
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-07-15 17:05:57 +02:00
Code Issues Packages Projects Releases Wiki Activity

Commit Graph

  • 2d2e1030e3 docs : update opencl ops (#17904) lhez 2025-12-10 06:20:00 -08:00
  • c02654eb7d graph : make the compute graph constant with respect to active samplers Georgi Gerganov 2025-12-10 15:54:33 +02:00
  • 0ecee8be37 server : reconnect the backend_sampling setting in the WebUI Georgi Gerganov 2025-12-10 15:42:02 +02:00
  • 81cb5783c8 Merge branch 'master' into HEAD Georgi Gerganov 2025-12-10 13:41:32 +02:00
  • 17f7f4baad CUDA: fix unpadded strides in MMA FA kernel (#17891) b7345 Johannes Gäßler 2025-12-10 12:39:56 +01:00
  • 9e79b0116e convert: allow using quantized Mistral weight (#17889) Xuan-Son Nguyen 2025-12-10 10:26:22 +01:00
  • 2e9eab80c2 fix softmax for iGPU (#17838) b7343 Neo Zhang Jianyu 2025-12-10 16:59:57 +08:00
  • 2fbe3b7bb7 common : add parser for ministral/mistral large 3/devstral 2 (#17713) b7342 Aldehir Rojas 2025-12-09 17:31:04 -06:00
  • 63391852b0 docs : update cpu and cuda ops (#17890) Sigbjørn Skjæret 2025-12-09 23:31:29 +01:00
  • 086a63e3a5 metal: SSM kernel improvements (#17876) b7340 Gabe Goodhart 2025-12-09 12:30:02 -07:00
  • b63509262a Add DIAG for CUDA (#17873) b7339 Piotr Wilkin (ilintar) 2025-12-09 20:28:57 +01:00
  • 48f47565a7 docs: clarify that CPU support should be first (#17886) Johannes Gäßler 2025-12-09 20:10:36 +01:00
  • 6dc6614bf0 Disable cooperative groups for musa Oliver Simons 2025-12-09 19:09:52 +01:00
  • a25fda5290 Fix launch logic when supports_cooperative_launch=false Oliver Simons 2025-12-09 19:03:47 +01:00
  • 3f0594ad0b Try fixing HIP build errors by adding corresponding #defines Oliver Simons 2025-12-09 18:51:28 +01:00
  • 02e409a5be ggml : Provide macos-specific backtrace printing to avoid terminal death (#17869) b7337 Gabe Goodhart 2025-12-09 09:29:07 -07:00
  • 34b407b41c sampling : use host buffer type for inputs Georgi Gerganov 2025-12-09 17:53:17 +02:00
  • 92ff767918 llama : require backend samplers to be of type llama_sampler_chain Georgi Gerganov 2025-12-09 15:38:37 +02:00
  • 6b82eb7883 metal : print node names for debugging (#17882) b7336 Georgi Gerganov 2025-12-09 15:25:49 +02:00
  • 07003f1ffb Fix compiler warnings by casting const away Oliver Simons 2025-12-09 13:05:43 +01:00
  • 886c3668b5 Add TODOs to and adjust heuristics of row-wise soft_max in CUDA Oliver Simons 2025-12-09 12:55:30 +01:00
  • a84dfd3e10 CUDA: Add Cooperative-Groups-based parallelization of ncols in softmax Oliver Simons 2025-12-08 16:48:52 +01:00
  • 86a3f0fad8 ggml : allow fill node alloc inplace (#17870) b7335 Sigbjørn Skjæret 2025-12-09 12:23:47 +01:00
  • 63908b631a cmake: fix Mach-O current version number (#17877) b7334 Rhys-T 2025-12-09 06:17:41 -05:00
  • 42b12b5608 model : nit, DeepSeek V1 MoE is 16B and GigaChat is 20B (#12652) b7333 Sigbjørn Skjæret 2025-12-09 12:15:06 +01:00
  • 4e842d5120 console: allow using arrow left/right, home/end keys and history mode (#17836) b7332 Xuan-Son Nguyen 2025-12-09 11:53:59 +01:00
  • 7ab6f51b97 Revert "ggml : remove redundant src in ggml_cast" Georgi Gerganov 2025-12-09 12:52:59 +02:00
  • ca709e427b CANN: add support for partial RoPE and Vision mode (#17543) b7331 Chenguang Li 2025-12-09 17:53:23 +08:00
  • 2a615b27e4 ggml : remove redundant src in ggml_cast gg/cast-remove-src Georgi Gerganov 2025-12-09 10:58:06 +02:00
  • 9f6681c3a4 ggml-alloc : fix reuse-parent logic for misaligned sizes Georgi Gerganov 2025-12-09 11:13:44 +02:00
  • 62d1b0082d ggml : remove redundant src in ggml_cast Georgi Gerganov 2025-12-09 10:58:06 +02:00
  • d62b5804e1 metal : print node names for debugging Georgi Gerganov 2025-12-09 10:55:54 +02:00
  • 560ac16f7d server : handle unsupported cases Georgi Gerganov 2025-12-09 10:55:11 +02:00
  • 0cdce38a97 CUDA: fix FP16 overflow in tile FA kernel (#17875) b7330 Johannes Gäßler 2025-12-09 09:34:02 +01:00
  • e39502e74b llama : add token matching support to llama-grammar (#17816) b7329 Aldehir Rojas 2025-12-09 00:32:57 -06:00
  • 1d2a1ab73d model : support Rnj-1 (#17811) b7328 philip-essential 2025-12-08 19:49:03 -08:00
  • c8554b66e0 graph : use fill instead of scale_bias in grouped expert selection (#17867) b7327 Sigbjørn Skjæret 2025-12-08 21:29:59 +01:00
  • f3beb22b17 sampling : handle n_probs case Georgi Gerganov 2025-12-08 21:30:10 +02:00
  • 2fa51c19b0 model-conversion : add token ids to prompt token output [no ci] (#17863) Daniel Bevenius 2025-12-08 17:13:08 +01:00
  • 951520ddb0 server: delegate result_state creation to server_task (#17835) b7325 Xuan-Son Nguyen 2025-12-08 17:04:38 +01:00
  • 6d38db5dfe Merge branch 'master' into HEAD Georgi Gerganov 2025-12-08 17:55:24 +02:00
  • 68522c678d ci : support bfloat16 SYCL release package (#17855) b7324 Neo Zhang 2025-12-08 22:09:39 +08:00
  • f896d2c34f server: improve speed of speculative decoding (#17808) Xuan-Son Nguyen 2025-12-08 14:35:28 +01:00
  • e4e9c4329c Make graph_max_nodes vary by ubatch size (#17794) Piotr Wilkin (ilintar) 2025-12-08 14:32:41 +01:00
  • 636fc17a37 Fix Kimi-K2 tool-call parsing issues (#17376) hksdpc255 2025-12-09 00:32:04 +11:00
  • 51e0c2d917 cuda : add FILL op support (#17851) Jay Zenith 2025-12-08 05:10:12 -08:00
  • 37a4f63244 server : add development documentation (#17760) Xuan-Son Nguyen 2025-12-08 13:54:58 +01:00
  • 2bc96931d2 server : make cache_reuse configurable per request (#17858) b7318 Georgi Gerganov 2025-12-08 12:43:12 +02:00
  • 5814b4dce1 cuda: optimize SOLVE_TRI using registers and FMAF (#17703) b7317 wsbagnsv1 2025-12-08 10:41:08 +01:00
  • 79d61896d3 ggml-cpu: add ggml_thread_cpu_relax with Zihintpause support (#17784) b7316 ixgbe 2025-12-08 16:41:34 +08:00
  • 4d3726278b model: add llama 4 scaling for mistral-large (deepseek arch) (#17744) b7315 Xuan-Son Nguyen 2025-12-07 22:29:54 +01:00
  • 08f9d3cc1d Vulkan: improve mul_mat_vec_iq1_m (#16907) b7314 lovedheart 2025-12-07 18:40:42 +01:00
  • 72e3681073 sampling : fix top-p Georgi Gerganov 2025-12-07 17:11:50 +02:00
  • 42125f0e10 tests : check temp back to 0.0 Georgi Gerganov 2025-12-07 15:54:49 +02:00
  • 8ef5f900db cont : fixes Georgi Gerganov 2025-12-07 12:52:25 +02:00
  • 0a540f9abd ci : add windows-cuda 13.1 release (#17839) b7313 Sigbjørn Skjæret 2025-12-07 14:02:04 +01:00
  • 22577583a3 common : change --color to accept on/off/auto, default to auto (#17827) b7312 Sigbjørn Skjæret 2025-12-07 03:43:50 +01:00
  • d9e03db1e7 sycl: add missing BF16 conversion support for Intel oneAPI (#17780) b7311 Law Po Ying 2025-12-07 09:18:18 +08:00
  • db97837385 vulkan: perf_logger improvements (#17672) b7310 Jeff Bolz 2025-12-06 11:46:46 -06:00
  • 017761daf5 ggml-zendnn : add ZenDNN backend for AMD CPUs (#17690) Vishal Singh 2025-12-06 21:43:33 +05:30
  • 52258181da tests : fix memory leaks Georgi Gerganov 2025-12-06 17:11:15 +02:00
  • fdac9686f7 Merge branch 'master' into HEAD Georgi Gerganov 2025-12-06 16:55:33 +02:00
  • c42712b056 server: support multiple generations from one prompt (OAI "n" option) (#17775) Xuan-Son Nguyen 2025-12-06 15:54:38 +01:00
  • 30742a6ff5 sampling : expand support (wip) Georgi Gerganov 2025-12-05 22:02:48 +02:00
  • 09c7c50e64 ggml : add circular tiling support to pad, for Vulkan, CUDA, and CPU (used for making seamless textures) (#16985) b7307 Phylliida Dev 2025-12-06 06:07:02 -08:00
  • f334b79494 HIP: fix RDNA3 FP16/BF16 matrix multiplication (#17817) b7306 Johannes Gäßler 2025-12-06 13:45:36 +01:00
  • a28e3c7567 webui: Stop generation from chat sidebar (#17806) Aleksander Grygier 2025-12-06 13:29:15 +01:00
  • e31b5c55c3 webui: Fix context available value in Multi-model Router mode (#17804) Aleksander Grygier 2025-12-06 13:23:29 +01:00
  • 21f24f27a9 webui: Per-conversation system message with UI displaying, edition & branching (#17275) Aleksander Grygier 2025-12-06 13:19:05 +01:00
  • 7b43f55753 ggml : improve error handling for search path existence checks (#17653) b7302 Sky 2025-12-06 19:28:16 +08:00
  • 444f00b0ec llama : remove quantization sanity check (#17788) b7301 Daniel Bevenius 2025-12-06 12:26:20 +01:00
  • 2960eb2975 vulkan: Use one row per workgroup for f32 mmv (#17711) b7300 Jeff Bolz 2025-12-06 04:12:26 -06:00
  • dbc15a7967 convert: support Mistral 3 Large MoE (#17730) Xuan-Son Nguyen 2025-12-06 10:49:33 +01:00
  • c6c5e85979 vulkan: support solve_tri with larger N/K values (#17781) b7298 Jeff Bolz 2025-12-06 01:56:45 -06:00
  • 8e5f4987b1 contrib : stale PRs (#17803) Georgi Gerganov 2025-12-06 09:34:18 +02:00
  • 8ce774a102 metal : fix build(#17799) b7296 Georgi Gerganov 2025-12-06 09:33:59 +02:00
  • 67788f6846 vulkan: Replace deprecated VK_EXT_validation_features (#17637) Masato Nakasaka 2025-12-06 14:39:42 +09:00
  • d8c0a7b085 vulkan: Fix mismatch in TOPK_MOE unit test (#17541) Masato Nakasaka 2025-12-06 14:23:30 +09:00
  • 933414c0b6 vulkan: add more num_blocks instantiations in rms_norm (#17701) b7293 Jeff Bolz 2025-12-05 15:08:56 -06:00
  • a0f3897d53 vulkan: fix top_k bug when there are ties in the input (#17659) Jeff Bolz 2025-12-05 15:03:19 -06:00
  • 31436df5ae contrib : stale PRs gg/contrib-stale Georgi Gerganov 2025-12-05 22:49:15 +02:00
  • e15cd06a94 vulkan : support conv-2d with large output size (#17685) Acly 2025-12-05 21:46:39 +01:00
  • fd57b24c0f ggml webgpu: unary op suppport, code refactoring, ops support (#17764) Reese Levine 2025-12-05 12:25:51 -08:00
  • 6ab0d64960 vulkan: enable mmvq for q2_k on NVIDIA (#17675) Jeff Bolz 2025-12-05 14:21:57 -06:00
  • 93bb92664e vulkan: set all memory allocations to high priority (#17624) Jeff Bolz 2025-12-05 14:21:04 -06:00
  • 8160b38a5f rpc : fix alloc size logic (#17116) Georgi Gerganov 2025-12-05 19:39:04 +02:00
  • c41bde6fbd metal : add residency sets keep-alive heartbeat (#17766) Georgi Gerganov 2025-12-05 19:38:54 +02:00
  • e652566139 Readd cub::DeviceScan::InclusiveSum-based CumSum Oliver Simons 2025-12-05 16:15:31 +01:00
  • 7668999518 Merge branch 'master' into gpu-sampling Oliver Simons 2025-12-05 14:41:08 +01:00
  • dd11f6eb7b Add perf-tests for CUMSUM Oliver Simons 2025-12-05 13:54:44 +01:00
  • 6016d0bd41 HIP : fix RDNA4 build (#17792) b7285 Johannes Gäßler 2025-12-05 13:47:52 +01:00
  • cf74b1a8ec sampling : fix candidates logic Georgi Gerganov 2025-12-05 14:21:08 +02:00
  • 1be97831e4 fix: prevent segfault in tokenizer on highly repetitive input (#17786) Pascal 2025-12-05 12:52:23 +01:00
  • a6cfc212ed ci : fix winget workflow (#17790) Adrien Gallouët 2025-12-05 12:44:17 +01:00
  • 3a0d10533a Q4/Q8 Tiled Gemm Optimization. (#16999) shalinib-ibm 2025-12-05 17:11:51 +05:30
  • 6648989673 Add pwilkin to CODEOWNERS for chat files (#17789) Piotr Wilkin (ilintar) 2025-12-05 12:00:57 +01:00
  • e95d0bc8fd CUDA: fix FA VKQ accumulator overflow (#17746) Johannes Gäßler 2025-12-05 09:18:10 +01:00
  • 668ed76574 HIP: enable WMMA-MMQ INT kernels for RDNA 3 (#17576) Jiacheng (Jason) Chen 2025-12-05 03:17:37 -05:00
  • 03d9a77b85 ci : transform release binary root dir in tar to llama-bXXXX (#17773) b7278 Sigbjørn Skjæret 2025-12-05 01:50:19 +01:00
  • 3143a755c8 docs : update ops.md (Metal, BLAS) (#17768) Gabe Goodhart 2025-12-04 16:55:34 -07:00