diff --git a/docker/Dockerfile.ubuntu b/docker/Dockerfile.ubuntu index be294205..73ab8713 100644 --- a/docker/Dockerfile.ubuntu +++ b/docker/Dockerfile.ubuntu @@ -69,12 +69,13 @@ RUN if [ "$IMAGE_TYPE" = "gpu" ]; then \ pip install --no-cache-dir mcore-bridge -i https://pypi.org/simple/ -U && \ pip install --no-cache-dir pybind11 liger_kernel wandb swanlab nvitop pre-commit "transformers<5.9" "trl<1.0" "peft<0.20" huggingface-hub -U && \ pip install git+https://github.com/NVIDIA/TransformerEngine.git@stable --no-build-isolation; \ - pip install "flash-attn-4[cu13]==4.0.0b12"; \ + pip install "flash-attn-4[cu13]==4.0.0b15"; \ pip install git+https://github.com/deepseek-ai/DeepGEMM.git@v2.1.1.post3 --no-build-isolation; \ pip install -U flash-linear-attention --no-build-isolation; \ pip install -U git+https://github.com/Dao-AILab/causal-conv1d --no-build-isolation; \ pip install git+https://github.com/Dao-AILab/fast-hadamard-transform --no-build-isolation; \ pip install git+https://github.com/NVIDIA-NeMo/Emerging-Optimizers.git@v0.2.0; \ + pip uninstall tilelang -y; \ cd /tmp && GIT_LFS_SKIP_SMUDGE=1 git clone https://github.com/NVIDIA/apex && \ cd apex && pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./ && \ cd / && rm -fr /tmp/apex && pip cache purge; \ diff --git a/docker/build_image.py b/docker/build_image.py index fa334437..fc525bac 100644 --- a/docker/build_image.py +++ b/docker/build_image.py @@ -405,7 +405,7 @@ class LatestGPUImageBuilder(StableGPUImageBuilder): args.torchaudio_version = '2.11.0' args.torchvision_version = '0.26.0' if not args.vllm_version: - args.vllm_version = '0.20.1' + args.vllm_version = '0.22.0' if not args.base_image: # A mirrored image of nvidia/cuda:12.4.0-devel-ubuntu22.04 args.base_image = 'nvidia/cuda:13.0.3-cudnn-devel-ubuntu22.04'