xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 15:25:28 +08:00
Code Issues Packages Projects Releases Wiki Activity

Directly get max encoder len from VLLM config in V1 (#24866)

Browse Source 
Signed-off-by: Sugar-zsg <952242923@qq.com>
This commit is contained in:
Sugar 2025-09-17 01:52:31 +08:00 committed by GitHub
parent d593cf28fa
commit cd1f885bcf
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/attention/layers/cross_attention.py
View File

@ -14,7 +14,6 @@ from vllm.attention.layer import Attention
from vllm.attention.selector import get_attn_backend from vllm.attention.selector import get_attn_backend
from vllm.config import CacheConfig, VllmConfig from vllm.config import CacheConfig, VllmConfig
from vllm.logger import init_logger from vllm.logger import init_logger
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.utils import cdiv from vllm.utils import cdiv
from vllm.v1.attention.backends.utils import (CommonAttentionMetadata, from vllm.v1.attention.backends.utils import (CommonAttentionMetadata,
subclass_attention_backend) subclass_attention_backend)
@ -23,9 +22,13 @@ from vllm.v1.kv_cache_interface import CrossAttentionSpec
logger = init_logger(__name__) logger = init_logger(__name__)
def _get_max_encoder_len(vllm_config: VllmConfig) -> int: def _get_max_encoder_len(vllm_config: "VllmConfig") -> int:
return MULTIMODAL_REGISTRY.get_encdec_max_encoder_len( """Gets the max number of encoder input tokens from the config.
vllm_config.model_config) """
sc = vllm_config.scheduler_config
assert sc and isinstance(sc.max_num_encoder_input_tokens, int), \
"max_num_encoder_input_tokens must be int for enc-dec models"
return sc.max_num_encoder_input_tokens
def _get_cross_slot_mapping(encoder_seq_lens: np.ndarray, def _get_cross_slot_mapping(encoder_seq_lens: np.ndarray,