This commit is contained in:
hailin 2025-08-08 22:18:55 +08:00
parent afca830e53
commit 49d7fa3478
1 changed files with 15 additions and 0 deletions

View File

@ -166,6 +166,21 @@ fi
# =============================
log "Build & install DeepSpeed (training only, no inference ops)"
# 先把 libcurand 的开发包装上(提供无版本名 libcurand.so
set +e
${CONDA_BIN} install -y -c "nvidia/label/cuda-${CUDA_VERSION_PIN}.0" libcurand-dev
RC=$?
set -e
if [[ $RC -ne 0 ]]; then
log "nvidia channel 没装上 libcurand-dev回退到 conda-forge"
${CONDA_BIN} install -y -c conda-forge libcurand libcurand-dev
fi
# 链接器需要能找到这些库
export LD_LIBRARY_PATH="${CONDA_PREFIX}/lib64:${CONDA_PREFIX}/lib:${LD_LIBRARY_PATH:-}"
export LIBRARY_PATH="${CONDA_PREFIX}/lib64:${CONDA_PREFIX}/lib:${LIBRARY_PATH:-}"
export DS_BUILD_INFERENCE=0
export DS_BUILD_CUTLASS=0
export DS_BUILD_QUANTIZER=0