github actions

.github/workflows/build.yml

@@ -0,0 +1,28 @@
+name: Build
+
+on: [push]
+
+jobs:
+  build:
+    strategy:
+      fail-fast: false
+      matrix:
+        platform: [macos-latest, ubuntu-latest]
+        python-version: ["3.8"]
+    runs-on: ${{ matrix.platform }}
+    steps:
+      - uses: actions/checkout@v4
+      - name: Clone ggerganov/ggml
+        run: git clone https://github.com/ggerganov/ggml.git
+      - name: Install packages
+        run: pip install -r requirements.txt
+      - name: Saving model weights
+        run: python save_model_weights.py
+      - name: Creating build folder
+        run: mkdir build; cd build
+      - name: Cmake
+        run: cmake ..
+      - name: Make
+        run: make
+      - name: Try running model.cpp
+        run: ./model

CMakeLists.txt

@@ -6,4 +6,4 @@ add_subdirectory(ggml)
 set(TEST_TARGET model)
 add_executable(${TEST_TARGET} model.cpp)
 target_link_libraries(${TEST_TARGET} PRIVATE ggml)
-target_compile_features(${TEST_TARGET} PUBLIC cxx_std_14)
+target_compile_features(${TEST_TARGET} PUBLIC cxx_std_11)

README.md

@@ -29,6 +29,7 @@ cmake ..; make
 - [ ] Trying on real model
 - [ ] Adept template for real case usage
 - [ ] Writing comments
+- [ ] Add argparse for `model.cpp`
 - [ ] Support FP16
 - [ ] Quantization (?)
 

model.cpp

@@ -310,14 +310,15 @@ int main(void)
 {
     ggml_time_init();
 
+    // Create example tensor
+    std::vector<float> input = {1, 2, 3, 4, 5};
+
+    // Load model and run forward
     module model;
     load_model("../ggml-model.bin", model);
-
-    // perform computation in cpu
-    std::vector<float> input = {1, 2, 3, 4, 5};
     struct ggml_tensor *result = compute(model, input);
 
-    // get the result data pointer as a float array to print
+    // Printing
     std::vector<float> out_data(ggml_nelements(result));
     memcpy(out_data.data(), result->data, ggml_nbytes(result));
 

requirements.txt

@@ -0,0 +1,2 @@
+torch
+numpy