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

[v1][attention] Support Hybrid Allocator + FlashInfer (#21412)

Browse Source 
Signed-off-by: Chen Zhang <zhangch99@outlook.com>
This commit is contained in:
Chen Zhang
2025-07-29 18:45:29 -07:00
committed by GitHub
parent 0e36abf993
commit 555e7225bc
16 changed files with 85 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/v1/attention/backends/rocm_aiter_fa.py
View File

@@ -231,8 +231,8 @@ class AiterFlashAttentionMetadataBuilder(
AttentionMetadataBuilder[AiterFlashAttentionMetadata]):
full_cudagraph_supported: ClassVar[bool] = True
def __init__(self, kv_cache_spec: AttentionSpec, vllm_config: VllmConfig,
device: torch.device):
def __init__(self, kv_cache_spec: AttentionSpec, layer_names: list[str],
vllm_config: VllmConfig, device: torch.device):
self.vllm_config = vllm_config
self.model_config = vllm_config.model_config
self.parallel_config = vllm_config.parallel_config