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

[sgl] multilayereagleworkerv2 fix (#22954)

Browse Source
This commit is contained in:
Tarushii Goel
2026-04-20 16:22:16 -07:00
committed by GitHub
parent 57ecce9807
commit 28f3a2d8ed
1 changed files with 6 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/speculative/multi_layer_eagle_worker_v2.py
+6
View File
@@ -394,6 +394,9 @@ class MultiLayerEagleDraftWorker(BaseDraftWorker):
topk_p_list = [] topk_p_list = []
topk_index_list = [] topk_index_list = []
for step in range(self.speculative_num_steps): for step in range(self.speculative_num_steps):
forward_batch.req_to_token_pool = self.draft_runner_list[
step
].req_to_token_pool
output: ModelRunnerOutput = self.draft_runner_list[step].forward( output: ModelRunnerOutput = self.draft_runner_list[step].forward(
forward_batch forward_batch
) )
@@ -498,6 +501,9 @@ class MultiLayerEagleDraftWorker(BaseDraftWorker):
draft_logits_output.topk_index, draft_logits_output.topk_index,
) )
else: else:
forward_batch.req_to_token_pool = self.draft_runner_list[
step
].req_to_token_pool
draft_logits_output = self.draft_runner_list[step].forward( draft_logits_output = self.draft_runner_list[step].forward(
forward_batch, skip_attn_backend_init=True forward_batch, skip_attn_backend_init=True
) )