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

[V1][CUDA] Full cudagraph support for FlashInfer (#21367)

Browse Source
This commit is contained in:
fhl2000
2025-08-02 09:49:34 +08:00
committed by GitHub
parent 3654847db5
commit 23322431c8
8 changed files with 376 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/v1/attention/backends/triton_attn.py
View File

@@ -18,7 +18,8 @@ from vllm.config import VllmConfig
from vllm.logger import init_logger
from vllm.platforms import current_platform
from vllm.v1.attention.backends.flash_attn import FlashAttentionMetadata
from vllm.v1.attention.backends.utils import (AttentionMetadataBuilder,
from vllm.v1.attention.backends.utils import (AttentionCGSupport,
AttentionMetadataBuilder,
CommonAttentionMetadata)
from vllm.v1.kv_cache_interface import AttentionSpec
@@ -57,7 +58,8 @@ class TritonAttentionMetadata:
class TritonAttentionMetadataBuilder(
AttentionMetadataBuilder[TritonAttentionMetadata]):
full_cudagraph_supported: ClassVar[bool] = True
attn_cudagraph_support: ClassVar[AttentionCGSupport] = \
AttentionCGSupport.ALWAYS
def __init__(self, kv_cache_spec: AttentionSpec, layer_names: list[str],
vllm_config: VllmConfig, device: torch.device):