Revert "TMP : push artifacts"

This reverts commit 537b09e70f.

.github/workflows/build.yml

@@ -56,6 +56,7 @@ jobs:
           mkdir build
           cd build
           cmake .. \
+            -DCMAKE_BUILD_RPATH="@loader_path" \
             -DLLAMA_FATAL_WARNINGS=ON \
             -DLLAMA_CURL=ON \
             -DGGML_METAL_USE_BF16=ON \
@@ -120,6 +121,7 @@ jobs:
           # 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 \
+            -DCMAKE_BUILD_RPATH="@loader_path" \
             -DLLAMA_FATAL_WARNINGS=ON \
             -DLLAMA_CURL=ON \
             -DGGML_METAL=OFF \
@@ -160,8 +162,8 @@ jobs:
           path: llama-${{ steps.tag.outputs.name }}-bin-macos-x64.zip
           name: llama-bin-macos-x64.zip
 
-  ubuntu-latest-cmake:
-    runs-on: ubuntu-latest
+  ubuntu-cpu-cmake:
+    runs-on: ubuntu-22.04
 
     steps:
       - name: Clone
@@ -181,7 +183,10 @@ jobs:
         run: |
           mkdir build
           cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_CURL=ON -DGGML_RPC=ON
+          cmake .. \
+            -DLLAMA_FATAL_WARNINGS=ON \
+            -DLLAMA_CURL=ON \
+            -DGGML_RPC=ON
           cmake --build . --config Release -j $(nproc)
 
       - name: Test
@@ -256,7 +261,10 @@ jobs:
         run: |
           mkdir build
           cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
+          cmake .. \
+            -DLLAMA_FATAL_WARNINGS=ON \
+            -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
+            -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
           cmake --build . --config ${{ matrix.build_type }} -j $(nproc)
 
       - name: Build (no OpenMP)
@@ -265,7 +273,11 @@ jobs:
         run: |
           mkdir build
           cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DGGML_OPENMP=OFF
+          cmake .. \
+            -DLLAMA_FATAL_WARNINGS=ON \
+            -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
+            -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
+            -DGGML_OPENMP=OFF
           cmake --build . --config ${{ matrix.build_type }} -j $(nproc)
 
       - name: Test
@@ -295,7 +307,8 @@ jobs:
         run: |
           mkdir build
           cd build
-          cmake -DGGML_RPC=ON ..
+          cmake .. \
+            -DGGML_RPC=ON
           cmake --build . --config Release -j $(nproc)
 
       - name: Test
@@ -325,7 +338,8 @@ jobs:
         run: |
           mkdir build
           cd build
-          cmake -DGGML_VULKAN=ON ..
+          cmake .. \
+            -DGGML_VULKAN=ON
           cmake --build . --config Release -j $(nproc)
 
       - name: Test
@@ -352,13 +366,18 @@ jobs:
       - name: Build with native CMake HIP support
         id: cmake_build
         run: |
-          cmake -B build -S . -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" -DGGML_HIP=ON
+          cmake -B build -S . \
+            -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" \
+            -DGGML_HIP=ON
           cmake --build build --config Release -j $(nproc)
 
       - name: Build with legacy HIP support
         id: cmake_build_legacy_hip
         run: |
-          cmake -B build2 -S . -DCMAKE_C_COMPILER=hipcc -DCMAKE_CXX_COMPILER=hipcc -DGGML_HIP=ON
+          cmake -B build2 -S . \
+            -DCMAKE_C_COMPILER=hipcc \
+            -DCMAKE_CXX_COMPILER=hipcc \
+            -DGGML_HIP=ON
           cmake --build build2 --config Release -j $(nproc)
 
   ubuntu-22-cmake-musa:
@@ -379,7 +398,8 @@ jobs:
       - name: Build with native CMake MUSA support
         id: cmake_build
         run: |
-          cmake -B build -S . -DGGML_MUSA=ON
+          cmake -B build -S . \
+            -DGGML_MUSA=ON
           cmake --build build --config Release -j $(nproc)
 
   ubuntu-22-cmake-sycl:
@@ -420,7 +440,10 @@ jobs:
           source /opt/intel/oneapi/setvars.sh
           mkdir build
           cd build
-          cmake -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx ..
+          cmake .. \
+            -DGGML_SYCL=ON \
+            -DCMAKE_C_COMPILER=icx \
+            -DCMAKE_CXX_COMPILER=icpx
           cmake --build . --config Release -j $(nproc)
 
   ubuntu-22-cmake-sycl-fp16:
@@ -461,42 +484,13 @@ jobs:
           source /opt/intel/oneapi/setvars.sh
           mkdir build
           cd build
-          cmake -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON ..
+          cmake .. \
+            -DGGML_SYCL=ON \
+            -DCMAKE_C_COMPILER=icx \
+            -DCMAKE_CXX_COMPILER=icpx \
+            -DGGML_SYCL_F16=ON
           cmake --build . --config Release -j $(nproc)
 
-  # TODO: build with GGML_METAL=OFF because test-backend-ops fail on "Apple Paravirtual device" and I don't know
-  #       how to debug it.
-  #       ref: https://github.com/ggerganov/llama.cpp/actions/runs/7132125951/job/19422043567?pr=4359#step:5:6584
-  #       would be great if we fix these
-  macOS-latest-cmake:
-    runs-on: macos-latest
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Dependencies
-        id: depends
-        continue-on-error: true
-        run: |
-          brew update
-
-      - name: Build
-        id: cmake_build
-        run: |
-          sysctl -a
-          mkdir build
-          cd build
-          cmake -DLLAMA_FATAL_WARNINGS=ON -DGGML_METAL=OFF ..
-          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          ctest -L main --verbose --timeout 900
-
   macOS-latest-cmake-ios:
     runs-on: macos-latest
 
@@ -827,7 +821,13 @@ jobs:
 
         - name: Build with CMake
           run: |
-            cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=89-real -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined -DLLAMA_FATAL_WARNINGS=ON
+            cmake -S . -B build -G Ninja \
+              -DCMAKE_BUILD_TYPE=Release \
+              -DCMAKE_CUDA_ARCHITECTURES=89-real \
+              -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined \
+              -DLLAMA_FATAL_WARNINGS=ON \
+              -DGGML_NATIVE=OFF \
+              -DGGML_CUDA=ON
             cmake --build build
 
   windows-2019-cmake-cuda:
@@ -916,7 +916,11 @@ jobs:
         shell: cmd
         run: |
           call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat"
-          cmake -S . -B build -G "Ninja Multi-Config" -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DGGML_RPC=ON
+          cmake -S . -B build -G "Ninja Multi-Config" \
+            -DLLAMA_BUILD_SERVER=ON \
+            -DGGML_NATIVE=OFF \
+            -DGGML_CUDA=ON \
+            -DGGML_RPC=ON
           set /A NINJA_JOBS=%NUMBER_OF_PROCESSORS%-1
           cmake --build build --config Release -j %NINJA_JOBS% -t ggml
           cmake --build build --config Release
@@ -1201,8 +1205,7 @@ jobs:
     runs-on: ubuntu-latest
 
     needs:
-      - ubuntu-latest-cmake
-      - macOS-latest-cmake
+      - ubuntu-cpu-cmake
       - windows-latest-cmake
       - windows-2019-cmake-cuda
       - windows-latest-cmake-hip-release

CMakeLists.txt

@@ -16,6 +16,7 @@ endif()
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")
 
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
+set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
 
 if (CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
     set(LLAMA_STANDALONE ON)

examples/run/README.md

@@ -3,11 +3,10 @@
 The purpose of this example is to demonstrate a minimal usage of llama.cpp for running models.
 
 ```bash
-llama-run granite-code
+llama-run granite3-moe
 ```
 
 ```bash
-llama-run -h
 Description:
   Runs a llm
 
@@ -17,7 +16,7 @@ Usage:
 Options:
   -c, --context-size <value>
       Context size (default: 2048)
-  -n, --ngl <value>
+  -n, -ngl, --ngl <value>
       Number of GPU layers (default: 0)
   --temp <value>
       Temperature (default: 0.8)

examples/run/run.cpp

@@ -147,7 +147,8 @@ class Opt {
                 if (handle_option_with_value(argc, argv, i, context_size) == 1) {
                     return 1;
                 }
-            } else if (options_parsing && (strcmp(argv[i], "-n") == 0 || strcmp(argv[i], "--ngl") == 0)) {
+            } else if (options_parsing &&
+                       (strcmp(argv[i], "-n") == 0 || strcmp(argv[i], "-ngl") == 0 || strcmp(argv[i], "--ngl") == 0)) {
                 if (handle_option_with_value(argc, argv, i, ngl) == 1) {
                     return 1;
                 }
@@ -194,7 +195,7 @@ class Opt {
             "Options:\n"
             "  -c, --context-size <value>\n"
             "      Context size (default: %d)\n"
-            "  -n, --ngl <value>\n"
+            "  -n, -ngl, --ngl <value>\n"
             "      Number of GPU layers (default: %d)\n"
             "  --temp <value>\n"
             "      Temperature (default: %.1f)\n"

examples/server/webui/index.html

@@ -141,6 +141,7 @@
               :msg="pendingMsg"
               :key="pendingMsg.id"
               :is-generating="isGenerating"
+              :show-thought-in-progress="config.showThoughtInProgress"
               :edit-user-msg-and-regenerate="() => {}"
               :regenerate-msg="() => {}"></message-bubble>
           </div>
@@ -202,6 +203,20 @@
               </template>
             </div>
           </details>
+          <!-- Section: Reasoning models -->
+          <details class="collapse collapse-arrow bg-base-200 mb-2 overflow-visible">
+            <summary class="collapse-title font-bold">Reasoning models</summary>
+            <div class="collapse-content">
+              <div class="flex flex-row items-center mb-2">
+                <input type="checkbox" class="checkbox" v-model="config.showThoughtInProgress" />
+                <span class="ml-4">Expand though process by default for generating message</span>
+              </div>
+              <div class="flex flex-row items-center mb-2">
+                <input type="checkbox" class="checkbox" v-model="config.excludeThoughtOnReq" />
+                <span class="ml-4">Exclude thought process when sending request to API (Recommended for DeepSeek-R1)</span>
+              </div>
+            </div>
+          </details>
           <!-- Section: Advanced config -->
           <details class="collapse collapse-arrow bg-base-200 mb-2 overflow-visible">
             <summary class="collapse-title font-bold">Advanced config</summary>
@@ -261,7 +276,17 @@
           <span v-if="msg.content === null" class="loading loading-dots loading-md"></span>
           <!-- render message as markdown -->
           <div v-else dir="auto">
-            <vue-markdown :source="msg.content"></vue-markdown>
+            <details v-if="msg.role === 'assistant' && splitMsgContent.cot" class="collapse bg-base-200 collapse-arrow mb-4" :open="splitMsgContent.isThinking && showThoughtInProgress">
+              <summary class="collapse-title">
+                <span v-if="splitMsgContent.isThinking">
+                  <span v-if="isGenerating" class="loading loading-spinner loading-md mr-2" style="vertical-align: middle;"></span>
+                  <b>Thinking</b>
+                </span>
+                <b v-else>Thought Process</b>
+              </summary>
+              <vue-markdown :source="splitMsgContent.cot" dir="auto" class="collapse-content"></vue-markdown>
+            </details>
+            <vue-markdown :source="splitMsgContent.content"></vue-markdown>
           </div>
           <!-- render timings if enabled -->
           <div class="dropdown dropdown-hover dropdown-top mt-2" v-if="timings && config.showTokensPerSecond">

examples/server/webui/src/main.js

@@ -17,6 +17,11 @@ import { asyncIterator } from '@sec-ant/readable-stream/ponyfill/asyncIterator';
 
 const isDev = import.meta.env.MODE === 'development';
 
+// types
+/** @typedef {{ id: number, role: 'user' | 'assistant', content: string, timings: any }} Message */
+/** @typedef {{ role: 'user' | 'assistant', content: string }} APIMessage */
+/** @typedef {{ id: string, lastModified: number, messages: Array<Message> }} Conversation */
+
 // utility functions
 const isString = (x) => !!x.toLowerCase;
 const isBoolean = (x) => x === true || x === false;
@@ -50,6 +55,8 @@ const CONFIG_DEFAULT = {
   apiKey: '',
   systemMessage: 'You are a helpful assistant.',
   showTokensPerSecond: false,
+  showThoughtInProgress: false,
+  excludeThoughtOnReq: true,
   // make sure these default values are in sync with `common.h`
   samplers: 'edkypmxt',
   temperature: 0.8,
@@ -172,6 +179,7 @@ const MessageBubble = defineComponent({
     config: Object,
     msg: Object,
     isGenerating: Boolean,
+    showThoughtInProgress: Boolean,
     editUserMsgAndRegenerate: Function,
     regenerateMsg: Function,
   },
@@ -188,7 +196,31 @@ const MessageBubble = defineComponent({
         prompt_per_second: this.msg.timings.prompt_n / (this.msg.timings.prompt_ms / 1000),
         predicted_per_second: this.msg.timings.predicted_n / (this.msg.timings.predicted_ms / 1000),
       };
-    }
+    },
+    splitMsgContent() {
+      const content = this.msg.content;
+      if (this.msg.role !== 'assistant') {
+        return { content };
+      }
+      let actualContent = '';
+      let cot = '';
+      let isThinking = false;
+      let thinkSplit = content.split('<think>', 2);
+      actualContent += thinkSplit[0];
+      while (thinkSplit[1] !== undefined) {
+        // <think> tag found
+        thinkSplit = thinkSplit[1].split('</think>', 2);
+        cot += thinkSplit[0];
+        isThinking = true;
+        if (thinkSplit[1] !== undefined) {
+          // </think> closing tag found
+          isThinking = false;
+          thinkSplit = thinkSplit[1].split('<think>', 2);
+          actualContent += thinkSplit[0];
+        }
+      }
+      return { content: actualContent, cot, isThinking };
+    },
   },
   methods: {
     copyMsg() {
@@ -208,7 +240,10 @@ const MessageBubble = defineComponent({
 // format: { [convId]: { id: string, lastModified: number, messages: [...] } }
 // convId is a string prefixed with 'conv-'
 const StorageUtils = {
-  // manage conversations
+  /**
+   * manage conversations
+   * @returns {Array<Conversation>}
+   */
   getAllConversations() {
     const res = [];
     for (const key in localStorage) {
@@ -219,11 +254,19 @@ const StorageUtils = {
     res.sort((a, b) => b.lastModified - a.lastModified);
     return res;
   },
-  // can return null if convId does not exist
+  /**
+   * can return null if convId does not exist
+   * @param {string} convId
+   * @returns {Conversation | null}
+   */
   getOneConversation(convId) {
     return JSON.parse(localStorage.getItem(convId) || 'null');
   },
-  // if convId does not exist, create one
+  /**
+   * if convId does not exist, create one
+   * @param {string} convId
+   * @param {Message} msg
+   */
   appendMsg(convId, msg) {
     if (msg.content === null) return;
     const conv = StorageUtils.getOneConversation(convId) || {
@@ -235,12 +278,24 @@ const StorageUtils = {
     conv.lastModified = Date.now();
     localStorage.setItem(convId, JSON.stringify(conv));
   },
+  /**
+   * Get new conversation id
+   * @returns {string}
+   */
   getNewConvId() {
     return `conv-${Date.now()}`;
   },
+  /**
+   * remove conversation by id
+   * @param {string} convId
+   */
   remove(convId) {
     localStorage.removeItem(convId);
   },
+  /**
+   * remove all conversations
+   * @param {string} convId
+   */
   filterAndKeepMsgs(convId, predicate) {
     const conv = StorageUtils.getOneConversation(convId);
     if (!conv) return;
@@ -248,6 +303,11 @@ const StorageUtils = {
     conv.lastModified = Date.now();
     localStorage.setItem(convId, JSON.stringify(conv));
   },
+  /**
+   * remove last message from conversation
+   * @param {string} convId
+   * @returns {Message | undefined}
+   */
   popMsg(convId) {
     const conv = StorageUtils.getOneConversation(convId);
     if (!conv) return;
@@ -322,10 +382,12 @@ const mainApp = createApp({
   data() {
     return {
       conversations: StorageUtils.getAllConversations(),
-      messages: [], // { id: number, role: 'user' | 'assistant', content: string }
+      /** @type {Array<Message>} */
+      messages: [],
       viewingConvId: StorageUtils.getNewConvId(),
       inputMsg: '',
       isGenerating: false,
+      /** @type {Array<Message> | null} */
       pendingMsg: null, // the on-going message from assistant
       stopGeneration: () => {},
       selectedTheme: StorageUtils.getTheme(),
@@ -333,6 +395,7 @@ const mainApp = createApp({
       showConfigDialog: false,
       // const
       themes: THEMES,
+      /** @type {CONFIG_DEFAULT} */
       configDefault: {...CONFIG_DEFAULT},
       configInfo: {...CONFIG_INFO},
       isDev,
@@ -425,42 +488,50 @@ const mainApp = createApp({
       this.isGenerating = true;
 
       try {
+        /** @type {CONFIG_DEFAULT} */
+        const config = this.config;
         const abortController = new AbortController();
         this.stopGeneration = () => abortController.abort();
+        /** @type {Array<APIMessage>} */
+        let messages = [
+          { role: 'system', content: config.systemMessage },
+          ...normalizeMsgsForAPI(this.messages),
+        ];
+        if (config.excludeThoughtOnReq) {
+          messages = filterThoughtFromMsgs(messages);
+        }
+        if (isDev) console.log({messages});
         const params = {
-          messages: [
-            { role: 'system', content: this.config.systemMessage },
-            ...this.messages,
-          ],
+          messages,
           stream: true,
           cache_prompt: true,
-          samplers: this.config.samplers,
-          temperature: this.config.temperature,
-          dynatemp_range: this.config.dynatemp_range,
-          dynatemp_exponent: this.config.dynatemp_exponent,
-          top_k: this.config.top_k,
-          top_p: this.config.top_p,
-          min_p: this.config.min_p,
-          typical_p: this.config.typical_p,
-          xtc_probability: this.config.xtc_probability,
-          xtc_threshold: this.config.xtc_threshold,
-          repeat_last_n: this.config.repeat_last_n,
-          repeat_penalty: this.config.repeat_penalty,
-          presence_penalty: this.config.presence_penalty,
-          frequency_penalty: this.config.frequency_penalty,
-          dry_multiplier: this.config.dry_multiplier,
-          dry_base: this.config.dry_base,
-          dry_allowed_length: this.config.dry_allowed_length,
-          dry_penalty_last_n: this.config.dry_penalty_last_n,
-          max_tokens: this.config.max_tokens,
-          timings_per_token: !!this.config.showTokensPerSecond,
-          ...(this.config.custom.length ? JSON.parse(this.config.custom) : {}),
+          samplers: config.samplers,
+          temperature: config.temperature,
+          dynatemp_range: config.dynatemp_range,
+          dynatemp_exponent: config.dynatemp_exponent,
+          top_k: config.top_k,
+          top_p: config.top_p,
+          min_p: config.min_p,
+          typical_p: config.typical_p,
+          xtc_probability: config.xtc_probability,
+          xtc_threshold: config.xtc_threshold,
+          repeat_last_n: config.repeat_last_n,
+          repeat_penalty: config.repeat_penalty,
+          presence_penalty: config.presence_penalty,
+          frequency_penalty: config.frequency_penalty,
+          dry_multiplier: config.dry_multiplier,
+          dry_base: config.dry_base,
+          dry_allowed_length: config.dry_allowed_length,
+          dry_penalty_last_n: config.dry_penalty_last_n,
+          max_tokens: config.max_tokens,
+          timings_per_token: !!config.showTokensPerSecond,
+          ...(config.custom.length ? JSON.parse(config.custom) : {}),
         };
         const chunks = sendSSEPostRequest(`${BASE_URL}/v1/chat/completions`, {
           method: 'POST',
           headers: {
             'Content-Type': 'application/json',
-            ...(this.config.apiKey ? {'Authorization': `Bearer ${this.config.apiKey}`} : {})
+            ...(config.apiKey ? {'Authorization': `Bearer ${config.apiKey}`} : {})
           },
           body: JSON.stringify(params),
           signal: abortController.signal,
@@ -477,7 +548,7 @@ const mainApp = createApp({
             };
           }
           const timings = chunk.timings;
-          if (timings && this.config.showTokensPerSecond) {
+          if (timings && config.showTokensPerSecond) {
             // only extract what's really needed, to save some space
             this.pendingMsg.timings = {
               prompt_n: timings.prompt_n,
@@ -598,3 +669,33 @@ try {
     <button class="btn" onClick="localStorage.clear(); window.location.reload();">Clear localStorage</button>
   </div>`;
 }
+
+/**
+ * filter out redundant fields upon sending to API
+ * @param {Array<APIMessage>} messages
+ * @returns {Array<APIMessage>}
+ */
+function normalizeMsgsForAPI(messages) {
+  return messages.map((msg) => {
+    return {
+      role: msg.role,
+      content: msg.content,
+    };
+  });
+}
+
+/**
+ * recommended for DeepsSeek-R1, filter out content between <think> and </think> tags
+ * @param {Array<APIMessage>} messages
+ * @returns {Array<APIMessage>}
+ */
+function filterThoughtFromMsgs(messages) {
+  return messages.map((msg) => {
+    return {
+      role: msg.role,
+      content: msg.role === 'assistant'
+        ? msg.content.split('</think>').at(-1).trim()
+        : msg.content,
+    };
+  });
+}

tests/test-backend-ops.cpp

@@ -3909,17 +3909,10 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
     test_cases.emplace_back(new test_gla(GGML_TYPE_F32, 32, 64, 32, 4));
     test_cases.emplace_back(new test_gla(GGML_TYPE_F32, 32, 64, 128, 4));
 
-    for (int i = 1; i < 9; ++i) {
-        test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16,    GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
-        test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q4_0,   GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
-        test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q4_1,   GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
-        test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q5_0,   GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
-        test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q5_1,   GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
-        test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q8_0,   GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
-        test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q4_K,   GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
-        test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q5_K,   GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
-        test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q6_K,   GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
-        test_cases.emplace_back(new test_mul_mat(GGML_TYPE_IQ4_NL, GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
+    for (ggml_type type_a : all_types) {
+        for (int i = 1; i < 10; ++i) {
+            test_cases.emplace_back(new test_mul_mat(type_a,    GGML_TYPE_F32, 16,  i, 256, { 1,  1}, {1, 1}));
+        }
     }
 
 #if 1