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

Reduce one step decode for draft model. (#11561)

Browse Source
This commit is contained in:
Liangsheng Yin
2025-10-14 23:52:04 +08:00
committed by GitHub
parent 56222658ec
commit 5ea96ac7cc
10 changed files with 253 additions and 211 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/attention/flashattention_backend.py
+2 -2
View File
@@ -2320,7 +2320,7 @@ class FlashAttentionMultiStepBackend:
self.topk = topk
self.speculative_num_steps = speculative_num_steps
self.attn_backends = []
for i in range(self.speculative_num_steps):
for i in range(self.speculative_num_steps - 1):
self.attn_backends.append(
FlashAttentionBackend(
model_runner,
@@ -2335,7 +2335,7 @@ class FlashAttentionMultiStepBackend:
self.attn_backends[i].init_forward_metadata(forward_batch)
def init_cuda_graph_state(self, max_bs: int, max_num_tokens: int):
for i in range(self.speculative_num_steps):
for i in range(self.speculative_num_steps - 1):
self.attn_backends[i].init_cuda_graph_state(max_bs, max_num_tokens)
def init_forward_metadata_capture_cuda_graph(