| |
| set -euo pipefail |
|
|
| VENV="${CLEF_VENV:-/workspace/clef-venv}" |
| CUDA_HOME="${CUDA_HOME:-/usr/local/cuda}" |
| CUDA_HOME="$(readlink -f "${CUDA_HOME}")" |
|
|
| export CUDA_HOME |
| export PATH="${CUDA_HOME}/bin:${PATH}" |
| export CPATH="${CUDA_HOME}/targets/x86_64-linux/include${CPATH:+:${CPATH}}" |
| export LIBRARY_PATH="${CUDA_HOME}/targets/x86_64-linux/lib/stubs${LIBRARY_PATH:+:${LIBRARY_PATH}}" |
| export TRITON_PTXAS_PATH="${TRITON_PTXAS_PATH:-${CUDA_HOME}/bin/ptxas}" |
| export TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-9.0}" |
| export MAX_JOBS="${MAX_JOBS:-8}" |
|
|
| python -m venv --system-site-packages "${VENV}" |
| "${VENV}/bin/python" -m pip install --upgrade pip |
| "${VENV}/bin/python" -m pip install --upgrade --no-cache-dir \ |
| transformers==5.12.1 \ |
| huggingface_hub \ |
| safetensors \ |
| pillow \ |
| accelerate \ |
| nvidia-cutlass-dsl==4.8.0 \ |
| apache-tvm-ffi==0.1.14.post1 \ |
| flash-linear-attention==0.5.2 |
| "${VENV}/bin/python" -m pip install --no-cache-dir --no-build-isolation \ |
| causal-conv1d==1.7.0 |
|
|
| if [[ "${CLEF_INSTALL_INT8:-0}" == "1" ]]; then |
| "${VENV}/bin/python" -m pip install --no-cache-dir --no-deps comfy-kitchen==0.2.31 |
| fi |
|
|
| "${VENV}/bin/python" - <<'PY' |
| import torch |
| from transformers.utils.import_utils import ( |
| is_causal_conv1d_available, |
| is_flash_linear_attention_available, |
| ) |
|
|
| print("torch", torch.__version__, torch.version.cuda) |
| print("flash_linear_attention", is_flash_linear_attention_available()) |
| print("causal_conv1d", is_causal_conv1d_available()) |
| PY |
|
|
