biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Core] Deprecate xformers (#29262)

Browse Source 
Signed-off-by: Roger Wang <hey@rogerw.io>
This commit is contained in:
Roger Wang
2025-11-23 20:18:55 -08:00
committed by GitHub
parent 5253f4276f
commit 0ff70821c9
31 changed files with 77 additions and 963 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
docs/contributing/ci/update_pytorch_version.md
View File

@@ -98,21 +98,6 @@ to warm it up so that future builds are faster.
<img width="60%" alt="Buildkite new build popup" src="https://github.com/user-attachments/assets/a8ff0fcd-76e0-4e91-b72f-014e3fdb6b94">
</p>
## Update dependencies
Several vLLM dependencies like xFormers depend on PyTorch and need
to be updated accordingly. Rather than waiting for all of them to publish new
releases (which would take too much time), they can be built from
source to unblock the update process.
### xFormers
```bash
export TORCH_CUDA_ARCH_LIST='7.5 8.0+PTX 9.0a'
MAX_JOBS=16 uv pip install --system \
--no-build-isolation "git+https://github.com/facebookresearch/xformers@v0.0.32.post2"
```
## Update all the different vLLM platforms
Rather than attempting to update all vLLM platforms in a single pull request, it's more manageable

2
docs/getting_started/quickstart.md
View File

@@ -283,7 +283,7 @@ Currently, vLLM supports multiple backends for efficient Attention computation a
If desired, you can also manually set the backend of your choice by configuring the environment variable `VLLM_ATTENTION_BACKEND` to one of the following options:
- On NVIDIA CUDA: `FLASH_ATTN`, `FLASHINFER` or `XFORMERS`.
- On NVIDIA CUDA: `FLASH_ATTN` or `FLASHINFER`.
- On AMD ROCm: `TRITON_ATTN`, `ROCM_ATTN`, `ROCM_AITER_FA` or `ROCM_AITER_UNIFIED_ATTN`.
For AMD ROCm, you can further control the specific Attention implementation using the following variables: