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

make trtllm attn backend's init_forward_metadat non blocking (#13802)

Browse Source
This commit is contained in:
yinghui
2025-11-23 13:35:40 -08:00
committed by GitHub
parent 2892265d4c
commit 18403f6bfe
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/attention/trtllm_mla_backend.py
+3 -1
View File
@@ -564,7 +564,9 @@ class TRTLLMMLABackend(FlashInferMLAAttnBackend):
seq_lens = forward_batch.seq_lens - forward_batch.extend_prefix_lens seq_lens = forward_batch.seq_lens - forward_batch.extend_prefix_lens
cum_seq_lens_q = torch.cat( cum_seq_lens_q = torch.cat(
( (
torch.tensor([0], device=forward_batch.seq_lens.device), torch.zeros(
1, dtype=torch.int32, device=forward_batch.seq_lens.device
),
torch.cumsum(seq_lens, dim=0), torch.cumsum(seq_lens, dim=0),
) )
).int() ).int()