frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Misc] Add check for dual_chunk_attention (#24070)

Browse Source 
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
This commit is contained in:
Jiangyun Zhu
2025-09-03 12:19:14 +08:00
committed by GitHub
parent 02d411fdb2
commit e81d4e69c1
1 changed files with 6 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/config/__init__.py
View File

@ -49,7 +49,8 @@ from vllm.transformers_utils.config import (
try_get_tokenizer_config, uses_mrope)
from vllm.transformers_utils.s3_utils import S3Model
from vllm.transformers_utils.utils import is_s3, maybe_model_redirect
from vllm.utils import (DEFAULT_MAX_NUM_BATCHED_TOKENS, LayerBlockType,
from vllm.utils import (DEFAULT_MAX_NUM_BATCHED_TOKENS,
STR_DUAL_CHUNK_FLASH_ATTN_VAL, LayerBlockType,
LazyLoader, common_broadcastable_dtype, random_uuid)
if TYPE_CHECKING:
@ -1304,6 +1305,10 @@ class ModelConfig:
self.hf_config.dual_chunk_attention_config[
"sparse_attention_enabled"] = True
if envs.VLLM_ATTENTION_BACKEND != STR_DUAL_CHUNK_FLASH_ATTN_VAL:
raise ValueError("please set VLLM_ATTENTION_BACKEND to "
f"{STR_DUAL_CHUNK_FLASH_ATTN_VAL}")
def verify_async_output_proc(self, parallel_config, speculative_config,
device_config) -> None:
if not self.use_async_output_proc: