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

Signed-off-by: Chen Zhang <zhangch99@outlook.com>
This commit is contained in:
Chen Zhang
2025-07-30 01:45:29 +00:00
committed by GitHub
parent 0e36abf993
commit 555e7225bc
16 changed files with 85 additions and 57 deletions
+1
View File
@@ -305,6 +305,7 @@ def test_propose(num_speculative_tokens):
_Backend.FLASH_ATTN_VLLM_V1)
attn_metadata_builder = attn_metadata_builder_cls(
kv_cache_spec=create_standard_kv_cache_spec(proposer.vllm_config),
layer_names=proposer.attn_layer_names,
vllm_config=proposer.vllm_config,
device=device,
)