xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-17 03:37:03 +08:00
Code Issues Packages Projects Releases Wiki Activity

Use --attention-backend flag instead of VLLM_ATTENTION_BACKEND env var

Browse Source 
Per reviewer feedback, the VLLM_ATTENTION_BACKEND environment variable
is being deprecated in favor of the --attention-backend CLI flag.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Signed-off-by: yurekami <yurekami@users.noreply.github.com>
This commit is contained in:
yurekami 2025-12-24 15:16:10 +09:00
parent 3c8358c328
commit 79e0db60ee
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/v1/worker/cp_utils.py
View File

@ -34,8 +34,8 @@ def check_attention_cp_compatibility(vllm_config: VllmConfig) -> None:
f"current attention backend "
f"'{layer_impl.__class__.__name__}'.\n\n"
f"To resolve this issue, try one of the following:\n"
f" 1. Use a different attention backend by setting:\n"
f" export VLLM_ATTENTION_BACKEND=<backend>\n"
f" 1. Use a different attention backend by specifying:\n"
f" --attention-backend <backend>\n"
f" 2. Set cp_kv_cache_interleave_size to 1\n"
f" 3. Disable speculative decoding"
)
@ -48,8 +48,8 @@ def check_attention_cp_compatibility(vllm_config: VllmConfig) -> None:
f"'{layer_impl.__class__.__name__}' does not support this "
f"feature.\n\n"
f"To resolve this issue, try one of the following:\n"
f" 1. Use a compatible attention backend by setting:\n"
f" export VLLM_ATTENTION_BACKEND=<backend>\n"
f" 1. Use a compatible attention backend by specifying:\n"
f" --attention-backend <backend>\n"
f" Compatible backends: FLASH_ATTN, FLASHINFER, "
f"TRITON_MLA, FLASH_MLA, FLASH_ATTN_MLA, CUTLASS_MLA\n"
f" 2. Disable DCP by removing the "
@ -66,8 +66,8 @@ def check_attention_cp_compatibility(vllm_config: VllmConfig) -> None:
f"'{layer_impl.__class__.__name__}' does not support this "
f"feature.\n\n"
f"To resolve this issue, try one of the following:\n"
f" 1. Use a compatible attention backend by setting:\n"
f" export VLLM_ATTENTION_BACKEND=<backend>\n"
f" 1. Use a compatible attention backend by specifying:\n"
f" --attention-backend <backend>\n"
f" 2. Disable PCP by removing the "
f"--prefill-context-parallel-size flag\n\n"
f"For more information, see:\n"