Update torchchat.py #26
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Compile-dtype main | ||
| on: | ||
| push: | ||
| branches: | ||
| - main | ||
| pull_request: | ||
| workflow_dispatch: | ||
| jobs: | ||
| run-tinystories: | ||
| strategy: | ||
| matrix: | ||
| runner: [ubuntu-latest, macos-14] | ||
| 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 | ||
| python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager | ||
| cat ./output_eager | ||
| python generate.py --dtype ${DTYPE} --compile --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled | ||
| cat ./output_compiled | ||
| python export.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so | ||
| python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti | ||
| cat ./output_aoti | ||
| 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 | ||
| python generate.py --dtype ${DTYPE} --compile --quant '{"embedding" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled | ||
| cat ./output_compiled | ||
| python export.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so | ||
| python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti | ||
| cat ./output_aoti | ||
| 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 | ||
| python generate.py --dtype ${DTYPE} --compile --quant '{"embedding" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled | ||
| cat ./output_compiled | ||
| python export.py --dtype ${DTYPE} --quant '{"embedding" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so | ||
| python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti | ||
| cat ./output_aoti | ||
| 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 | ||
| python generate.py --dtype ${DTYPE} --compile --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled | ||
| cat ./output_compiled | ||
| python export.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 0}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so | ||
| python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti | ||
| cat ./output_aoti | ||
| 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 | ||
| python generate.py --dtype ${DTYPE} --compile --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled | ||
| cat ./output_compiled | ||
| python export.py --dtype ${DTYPE} --quant '{"linear:int8" : {"bitwidth": 8, "groupsize": 8}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so | ||
| python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti | ||
| cat ./output_aoti | ||
| 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 | ||
| python generate.py --dtype ${DTYPE} --compile --quant '{"linear:int4" : {"groupsize": 32}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled | ||
| cat ./output_compiled | ||
| python export.py --dtype ${DTYPE} --quant '{"linear:int4" : {"groupsize": 32}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so | ||
| if [ $(uname -s) == Linux ]; then | ||
| echo "skipping INT4 groupwise quantization because AOTI fails" | ||
| else | ||
| python generate.py --dtype ${DTYPE} --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so > ./output_aoti | ||
| cat ./output_aoti | ||
| fi | ||
| echo "tests complete for ${DTYPE}" | ||
| done | ||
| echo "tests complete for all dtypes!" | ||
