Upgrade to vllm 0.17.0 corex v4.1 overlay
This commit is contained in:
@@ -16,8 +16,8 @@ class AttentionConfig:
|
||||
backend: AttentionBackendEnum | None = None
|
||||
"""Attention backend to use. If None, will be selected automatically."""
|
||||
|
||||
flash_attn_version: Literal[2, 3] | None = None
|
||||
"""Force vllm to use a specific flash-attention version (2 or 3).
|
||||
flash_attn_version: Literal[2, 3, 4] | None = None
|
||||
"""Force vllm to use a specific flash-attention version (2, 3, or 4).
|
||||
Only valid when using the flash-attention backend."""
|
||||
|
||||
use_prefill_decode_attention: bool = False
|
||||
|
||||
Reference in New Issue
Block a user