PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度
update llama-cpp-python installation for CUDA CI (#1907)
Sun, Xuehao786cc78e3b1c
diff --git a/.azure-pipelines/scripts/cuda_unit_test/run_cuda_ut.sh b/.azure-pipelines/scripts/cuda_unit_test/run_cuda_ut.shindex dd38a18c4..a5c0fd0b1 100644--- a/.azure-pipelines/scripts/cuda_unit_test/run_cuda_ut.sh+++ b/.azure-pipelines/scripts/cuda_unit_test/run_cuda_ut.sh@@ -63,7 +63,7 @@ function run_unit_test() {echo "##[group]set up UT env..."cd "${BUILD_SOURCESDIRECTORY}" || exit 1uv pip install torch==2.12.0 torchvision torchao --index-url https://download.pytorch.org/whl/cu130- uv pip install https://github.com/XuehaoSun/llama-cpp-python/releases/download/v0.3.23/llama_cpp_python-0.3.23-py3-none-linux_x86_64.whl+ uv pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu130uv pip install 'git+https://github.com/ggml-org/llama.cpp.git#subdirectory=gguf-py'uv pip install -r test/test_cuda/requirements.txtuv pip install -r test/test_cuda/requirements_diffusion.txt@@ -168,7 +168,6 @@ function run_unit_test_vllm() {rm -rf /root/.venvuv venv --python=3.12 /root/.venvuv pip install -U pytest-cov pytest-html- vllm_version=$(curl -s https://api.github.com/repos/vllm-project/vllm/releases/latest | jq -r .tag_name | sed 's/^v//')uv pip install -r test/test_cuda/requirements_vllm.txt \--extra-index-url https://download.pytorch.org/whl/cu130 \--index-strategy unsafe-best-matchdiff --git a/.azure-pipelines/scripts/ut/collect_result.py b/.azure-pipelines/scripts/ut/collect_result.pyindex 7795031bf..6c1ac95f3 100644--- a/.azure-pipelines/scripts/ut/collect_result.py+++ b/.azure-pipelines/scripts/ut/collect_result.py@@ -39,18 +39,21 @@ class LogAnalyzer:"unittest_",)+ # pytest test logic failures: test ran but assertion/expectation failedFAILURE_MARKERS = ("FAILED","== FAILURES =="," failures",- "Killed","AssertionError",- "Error:",- "core dumped",- "Segmentation fault",)+ # runtime/system errors: process crashed, setup/teardown failed, or unhandled exceptionERROR_MARKERS = (+ "Aborted",+ "Killed",+ "Segmentation fault",+ "core dumped",+ "Error:","ERROR:","== ERRORS =="," errors:",diff --git a/.azure-pipelines/scripts/ut/run_ut_cuda.sh b/.azure-pipelines/scripts/ut/run_ut_cuda.shindex c631b4de0..7391a537e 100644--- a/.azure-pipelines/scripts/ut/run_ut_cuda.sh+++ b/.azure-pipelines/scripts/ut/run_ut_cuda.sh@@ -92,7 +92,7 @@ function run_unit_test() {rm -rf .coverage* *.xml *.htmluv pip install torch==2.12.0 torchvision torchao --index-url https://download.pytorch.org/whl/cu126- uv pip install https://github.com/XuehaoSun/llama-cpp-python/releases/download/v0.3.23/llama_cpp_python-0.3.23+cu128-py3-none-linux_x86_64.whl+ uv pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu125uv pip install 'git+https://github.com/ggml-org/llama.cpp.git#subdirectory=gguf-py'uv pip install -r test_cuda/requirements.txtuv pip install -r test_cuda/requirements_diffusion.txt