minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Spec] trtllm mha supports overlap plan stream (#25925)

Browse Source
This commit is contained in:
Hanming Lu
2026-05-23 03:25:57 -07:00
committed by GitHub
parent cb7b57955d
commit a5a64a311a
1 changed files with 5 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/attention/trtllm_mha_backend.py
+5
View File
@@ -537,6 +537,11 @@ class TRTLLMHAAttnBackend(FlashInferAttnBackend):
self._copy_swa_page_table(metadata, page_indices, max_seq_pages) self._copy_swa_page_table(metadata, page_indices, max_seq_pages)
self.forward_metadata = metadata self.forward_metadata = metadata
def update_verify_buffers_to_fill_after_draft(
self, spec_info: SpecInput, cuda_graph_bs: Optional[int]
):
pass
def get_cuda_graph_seq_len_fill_value(self) -> int: def get_cuda_graph_seq_len_fill_value(self) -> int:
"""Get the fill value for sequence lengths in CUDA graph.""" """Get the fill value for sequence lengths in CUDA graph."""
return 1 return 1