@@ -1097,31 +1097,6 @@ jobs:
10971097 run : |
10981098 export TORCHCHAT_ROOT=${PWD}
10991099 echo "et-git-hash=$(cat ${TORCHCHAT_ROOT}/install/.pins/et-pin.txt)" >> "$GITHUB_ENV"
1100- - name : Load or install ET
1101- id : install-et
1102- uses : actions/cache@v4
1103- with :
1104- path : |
1105- ./et-build
1106- ./torchchat/utils/scripts
1107- key : et-build-${{runner.os}}-${{runner.arch}}-${{env.et-git-hash}}-${{ hashFiles('**/install_et.sh') }}
1108- - if : ${{ steps.install-et.outputs.cache-hit != 'true' }}
1109- continue-on-error : true
1110- run : |
1111- echo "Installing ExecuTorch"
1112- bash torchchat/utils/scripts/install_et.sh
1113- - name : Install ExecuTorch python
1114- run : |
1115- echo "Install ExecuTorch python"
1116- export TORCHCHAT_ROOT=$PWD
1117- export ET_BUILD_DIR="et-build"
1118- ENABLE_ET_PYBIND="${1:-true}"
1119- source "torchchat/utils/scripts/install_utils.sh"
1120- install_executorch_python_libs $ENABLE_ET_PYBIND
1121- - name : Install runner
1122- run : |
1123- echo "Installing runner"
1124- bash torchchat/utils/scripts/build_native.sh et link_torchao_ops
11251100 - name : Install runner AOTI
11261101 id : install-runner-aoti
11271102 run : |
@@ -1135,9 +1110,6 @@ jobs:
11351110 python torchchat.py generate stories110M --temperature 0 --prompt "${PRMT}" --device cpu --dtype float32 --quantize '{"linear:a8wxdq": {"bitwidth": 4, "groupsize": 256, "has_weight_zeros": false}}'
11361111 echo "Generate compile"
11371112 python torchchat.py generate stories110M --temperature 0 --prompt "${PRMT}" --device cpu --dtype float32 --quantize '{"linear:a8wxdq": {"bitwidth": 4, "groupsize": 256, "has_weight_zeros": false}}' --compile
1138- echo "Export and run ET (C++ runner)"
1139- python torchchat.py export stories110M --output-pte-path ./model.pte --dtype float32 --quantize '{"linear:a8wxdq": {"bitwidth": 4, "groupsize": 256, "has_weight_zeros": false}}'
1140- ./cmake-out/et_run ./model.pte -z ./tokenizer.model -t 0 -i "${PRMT}"
11411113 echo "Export and run AOTI (C++ runner)"
11421114 python torchchat.py export stories110M --output-dso-path ./model.so --dtype float32 --quantize '{"linear:a8wxdq": {"bitwidth": 4, "groupsize": 256, "has_weight_zeros": false}}'
11431115 ./cmake-out/aoti_run ./model.so -z ./tokenizer.model -t 0 -i "${PRMT}"
0 commit comments