model : add label for LiquidAI LFM2-2.6B model (#16204)

* model : add label for LiquidAI LFM2-2.6B model

HF link: [LiquidAI/LFM2-2.6B](https://huggingface.co/LiquidAI/LFM2-2.6B).

Support for GGUF conversion and inference is added in #14620.

However, due to similar `n_embd`, it identifies as a 1.2B model.
Fix the label by using `n_ff` to identify the model instead.

Output of `llama-bench`:
```
| model                          |       size |     params | backend    | threads |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --------------: | -------------------: |
| lfm2 1.2B F16                  |   2.18 GiB |     1.17 B | CPU        |      10 |           pp512 |        223.97 ± 5.32 |
| lfm2 2.6B F16                  |   4.79 GiB |     2.57 B | CPU        |      10 |           pp512 |         92.53 ± 4.14 |
| lfm2 350M F16                  | 676.25 MiB |   354.48 M | CPU        |      10 |           pp512 |       725.52 ± 11.70 |
| lfm2 700M F16                  |   1.38 GiB |   742.49 M | CPU        |      10 |           pp512 |       336.22 ± 12.93 |
```

* Update src/llama-model.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

.github/workflows/build.yml

@@ -1302,8 +1302,8 @@ jobs:
         run: |
           GG_BUILD_NO_BF16=1 GG_BUILD_EXTRA_TESTS_0=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
-  ggml-ci-x64-nvidia-v100-cuda:
-    runs-on: [self-hosted, Linux, X64, NVIDIA, V100]
+  ggml-ci-x64-nvidia-cuda:
+    runs-on: [self-hosted, Linux, X64, NVIDIA]
 
     steps:
       - name: Clone
@@ -1316,8 +1316,8 @@ jobs:
           nvidia-smi
           GG_BUILD_CUDA=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
-  ggml-ci-x64-nvidia-v100-vulkan:
-    runs-on: [self-hosted, Linux, X64, NVIDIA, V100]
+  ggml-ci-x64-nvidia-vulkan-cm:
+    runs-on: [self-hosted, Linux, X64, NVIDIA]
 
     steps:
       - name: Clone
@@ -1327,51 +1327,23 @@ jobs:
       - name: Test
         id: ggml-ci
         run: |
-          vulkaninfo
-          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
-
-  ggml-ci-x64-nvidia-t4-cuda:
-    runs-on: [self-hosted, Linux, X64, NVIDIA, T4]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          nvidia-smi
-          GG_BUILD_CUDA=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
-
-  ggml-ci-x64-nvidia-t4-vulkan:
-    runs-on: [self-hosted, Linux, X64, NVIDIA, T4]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          vulkaninfo
-          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
-
-  ggml-ci-x64-nvidia-t4-vulkan-coopmat1:
-    runs-on: [self-hosted, Linux, X64, NVIDIA, T4]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          vulkaninfo
+          vulkaninfo --summary
           GG_BUILD_VULKAN=1 GGML_VK_DISABLE_COOPMAT2=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
+  ggml-ci-x64-nvidia-vulkan-cm2:
+    runs-on: [self-hosted, Linux, X64, NVIDIA, COOPMAT2]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v4
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          vulkaninfo --summary
+          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
+
   ggml-ci-x64-cpu-amx:
     runs-on: [self-hosted, Linux, X64, CPU, AMX]
 
@@ -1385,31 +1357,33 @@ jobs:
         run: |
           bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
-  ggml-ci-x64-amd-v710-vulkan:
-    runs-on: [self-hosted, Linux, X64, AMD, V710]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
-
-  ggml-ci-x64-amd-v710-rocm:
-    runs-on: [self-hosted, Linux, X64, AMD, V710]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          GG_BUILD_ROCM=1 GG_BUILD_AMDGPU_TARGETS="gfx1101" bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
+#  ggml-ci-x64-amd-vulkan:
+#    runs-on: [self-hosted, Linux, X64, AMD]
+#
+#    steps:
+#      - name: Clone
+#        id: checkout
+#        uses: actions/checkout@v4
+#
+#      - name: Test
+#        id: ggml-ci
+#        run: |
+#          vulkaninfo --summary
+#          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
+#
+#  ggml-ci-x64-amd-rocm:
+#    runs-on: [self-hosted, Linux, X64, AMD]
+#
+#    steps:
+#      - name: Clone
+#        id: checkout
+#        uses: actions/checkout@v4
+#
+#      - name: Test
+#        id: ggml-ci
+#        run: |
+#          amd-smi static
+#          GG_BUILD_ROCM=1 GG_BUILD_AMDGPU_TARGETS="gfx1101" bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
   ggml-ci-mac-metal:
     runs-on: [self-hosted, macOS, ARM64]
@@ -1424,16 +1398,16 @@ jobs:
         run: |
           GG_BUILD_METAL=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
 
-# TODO: install vulkan drivers
-#  ggml-ci-mac-vulkan:
-#    runs-on: [self-hosted, macOS, ARM64]
-#
-#    steps:
-#      - name: Clone
-#        id: checkout
-#        uses: actions/checkout@v4
-#
-#      - name: Test
-#        id: ggml-ci
-#        run: |
-#          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+  ggml-ci-mac-vulkan:
+    runs-on: [self-hosted, macOS, ARM64]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v4
+
+      - name: Test
+        id: ggml-ci
+        run: |
+          vulkaninfo --summary
+          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp

CODEOWNERS

@@ -95,12 +95,12 @@
 /tools/tokenize/                        @ggerganov
 /tools/tts/                             @ggerganov
 /vendor/                                @ggerganov
-.clang-format                           @slaren
-.clang-tidy                             @slaren
-AUTHORS                                 @ggerganov
-CMakeLists.txt                          @ggerganov
-CONTRIBUTING.md                         @ggerganov
-LICENSE                                 @ggerganov
-README.md                               @ggerganov
-SECURITY.md                             @ggerganov
+/.clang-format                          @slaren
+/.clang-tidy                            @slaren
+/AUTHORS                                @ggerganov
+/CMakeLists.txt                         @ggerganov
+/CONTRIBUTING.md                        @ggerganov
+/LICENSE                                @ggerganov
+/README.md                              @ggerganov
+/SECURITY.md                            @ggerganov
 requirements*.txt                       @CISC

ci/run.sh

@@ -92,6 +92,12 @@ fi
 
 if [ ! -z ${GG_BUILD_VULKAN} ]; then
     CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_VULKAN=1"
+
+    # if on Mac, disable METAL
+    if [[ "$OSTYPE" == "darwin"* ]]; then
+        CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_METAL=OFF -DGGML_BLAS=OFF"
+    fi
+
 fi
 
 if [ ! -z ${GG_BUILD_WEBGPU} ]; then

common/common.cpp

@@ -14,6 +14,7 @@
 #include <climits>
 #include <cmath>
 #include <codecvt>
+#include <chrono>
 #include <cstdarg>
 #include <cstring>
 #include <ctime>

examples/model-conversion/README.md

@@ -105,12 +105,12 @@ new model, the model can be converted to GGUF format using the following command
 ### Inspecting the converted model
 The converted model can be inspected using the following command:
 ```console
-(venv) $ make inspect-converted-model
+(venv) $ make causal-inspect-converted-model
 ```
 
 ### Running the converted model
 ```console
-(venv) $ make run-converted-model
+(venv) $ make causal-run-converted-model
 ```
 
 ### Model logits verfication

examples/model-conversion/scripts/causal/compare-logits.py

@@ -48,7 +48,7 @@ def main():
         print(f"Error: Model file not found: {model_path}")
         sys.exit(1)
 
-    model_name = os.path.splitext(os.path.basename(model_path))[0]
+    model_name = os.path.basename(model_path)
     data_dir = Path("data")
 
     pytorch_file = data_dir / f"pytorch-{model_name}.bin"

examples/model-conversion/scripts/causal/run-org-model.py

@@ -193,7 +193,7 @@ print(f"Input text: {repr(prompt)}")
 print(f"Tokenized: {tokenizer.convert_ids_to_tokens(input_ids[0])}")
 
 with torch.no_grad():
-    outputs = model(input_ids)
+    outputs = model(input_ids.to(model.device))
     logits = outputs.logits
 
     # Extract logits for the last token (next token prediction)

examples/model-conversion/scripts/utils/check-nmse.py

@@ -67,7 +67,7 @@ def main():
     parser.add_argument('-m', '--model-path', required=True,  help='Path to the model directory')
     args = parser.parse_args()
 
-    model_name = os.path.splitext(os.path.basename(args.model_path))[0]
+    model_name = os.path.basename(args.model_path)
     data_dir = Path("data")
 
     pytorch_file = data_dir / f"pytorch-{model_name}.bin"

src/llama-model.cpp

@@ -66,6 +66,7 @@ const char * llm_type_name(llm_type type) {
         case LLM_TYPE_1_7B:          return "1.7B";
         case LLM_TYPE_1_8B:          return "1.8B";
         case LLM_TYPE_2B:            return "2B";
+        case LLM_TYPE_2_6B:          return "2.6B";
         case LLM_TYPE_2_8B:          return "2.8B";
         case LLM_TYPE_2_9B:          return "2.9B";
         case LLM_TYPE_3B:            return "3B";
@@ -1977,10 +1978,11 @@ void llama_model::load_hparams(llama_model_loader & ml) {
                 for (uint32_t il = 0; il < hparams.n_layer; ++il) {
                     hparams.recurrent_layer_arr[il] = hparams.n_head_kv(il) == 0;
                 }
-                switch (hparams.n_embd) {
-                    case 1024: type = LLM_TYPE_350M; break;
-                    case 1536: type = LLM_TYPE_700M; break;
-                    case 2048: type = LLM_TYPE_1_2B; break;
+                switch (hparams.n_ff()) {
+                    case  4608: type = LLM_TYPE_350M; break;
+                    case  6912: type = LLM_TYPE_700M; break;
+                    case  8192: type = LLM_TYPE_1_2B; break;
+                    case 10752: type = LLM_TYPE_2_6B; break;
                     default:   type = LLM_TYPE_UNKNOWN;
                 }
             } break;

src/llama-model.h

@@ -58,6 +58,7 @@ enum llm_type {
     LLM_TYPE_1_7B,
     LLM_TYPE_1_8B,
     LLM_TYPE_2B,
+    LLM_TYPE_2_6B,
     LLM_TYPE_2_8B,
     LLM_TYPE_2_9B,
     LLM_TYPE_3B,