Closed Eta0 closed 2 months ago
@Eta0 Build complete, success: https://github.com/coreweave/ml-containers/actions/runs/9524483529
Image: ghcr.io/coreweave/ml-containers/torch-extras:es-cuda-updates-7e3dcc7-base-cuda12.0.1-ubuntu20.04-torch2.3.0-vision0.18.0-audio2.3.0
@Eta0 Build complete, success: https://github.com/coreweave/ml-containers/actions/runs/9524483529
Image: ghcr.io/coreweave/ml-containers/torch-extras:es-cuda-updates-7e3dcc7-base-cuda11.8.0-ubuntu20.04-torch2.3.0-vision0.18.0-audio2.3.0
@Eta0 Build complete, success: https://github.com/coreweave/ml-containers/actions/runs/9524483529
Image: ghcr.io/coreweave/ml-containers/torch-extras:es-cuda-updates-7e3dcc7-base-cuda12.0.1-ubuntu22.04-torch2.3.0-vision0.18.0-audio2.3.0
Update CUDA,
xformers
, andflash_attn
CUDA
This change updates CUDA versions targeted in our CI for
ml-containers/torch
to include v12.3.2 and v12.4.1, and to simultaneously drop v12.1.1 to lighten the CI load.Some miscellaneous CI machinery had to change to account for the different naming schemes in the
nvidia/cuda
ancestor images since the introduction of cuDNN v9 alongside CUDA v12.3.torch-extras
ComponentsThis change additionally updates library components of
ml-containers/torch-extras
to include:xformers
v0.0.26.post1, up from v0.0.23.post1flash_attn
v2.5.9.post1, up from v2.5.8The update to
xformers
allows compatibility with newer versions of vLLM.