fix tests

.github/workflows/compile-bf16.yml → .github/workflows/compile-dtype.yml

@@ -11,7 +11,7 @@ jobs:
   run-tinystories:
     strategy:
       matrix:
-        runner: [ubuntu-latest, macos-14, macos-12]
+        runner: [ubuntu-latest, macos-14]
     runs-on: ${{matrix.runner}}
     steps:
       - name: Checkout repo
@@ -102,9 +102,6 @@ jobs:
             echo "******************************************"
             echo "******** INT4 group-wise quantized *******"
             echo "******************************************"
-              if [ ${DTYPE} == float16 ]; then
-              DTYPE=bfloat16
-            fi
 
             python generate.py --dtype ${DTYPE} --quant '{"linear:int4" : {"groupsize": 32}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
             cat ./output_eager

.github/workflows/eager-dtype.yml

@@ -0,0 +1,87 @@
+name: Compile-dtype main
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  run-tinystories:
+    strategy:
+      matrix:
+        runner: [macos-12]
+    runs-on: ${{matrix.runner}}
+    steps:
+      - name: Checkout repo
+        uses: actions/checkout@v2
+      - name: Setup Python
+        uses: actions/setup-python@v2
+        with:
+          python-version: 3.11
+      - name: Print machine info
+        run: |
+          uname -a
+          if [ $(uname -s) == Darwin ]; then
+            sysctl machdep.cpu.brand_string
+            sysctl machdep.cpu.core_count
+          fi
+      - name: Install requirements
+        run: |
+          pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu
+          pip install -r requirements.txt
+      - name: Download checkpoints
+        run: |
+          mkdir -p checkpoints/stories15M
+          pushd checkpoints/stories15M
+          wget https://huggingface.co/karpathy/tinyllamas/resolve/main/stories15M.pt
+          wget https://github.com/karpathy/llama2.c/raw/master/tokenizer.model
+          popd
+      - name: Run inference
+        run: |          
+          export MODEL_PATH=checkpoints/stories15M/stories15M.pt
+          export MODEL_NAME=stories15M
+          export MODEL_DIR=/tmp
+          for DTYPE in bfloat16 float16 float32; do
+            # if [ $(uname -s) == Darwin ]; then
+            #   export DTYPE=float16
+            # fi  
+            python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
+            cat ./output_eager
+  
+            echo "******************************************"
+            echo "******* Emb: channel-wise quantized ******"
+            echo "******************************************"
+            python generate.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
+            cat ./output_eager
+
+            echo "******************************************"
+            echo "******** Emb: group-wise quantized *******"
+            echo "******************************************"
+            python generate.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
+            cat ./output_eager
+
+            echo "******************************************"
+            echo "******* INT8 channel-wise quantized ******"
+            echo "******************************************"
+            python generate.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
+            cat ./output_eager
+
+            echo "******************************************"
+            echo "******** INT8 group-wise quantized *******"
+            echo "******************************************"
+            python generate.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
+            cat ./output_eager
+
+            echo "******************************************"
+            echo "******** INT4 group-wise quantized *******"
+            echo "******************************************"
+
+            python generate.py --dtype ${DTYPE} --quant '{"linear:int4" : {"groupsize": 32}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
+            cat ./output_eager
+
+            echo "tests complete for ${DTYPE}"
+            done
+
+          echo "tests complete for all dtypes!"