model: support Step-3.5-Flash (#18084)

Co-authored-by: ltd0924 <ltd0924@sina.com>
This commit is contained in:
Yuhao Yang
2026-02-03 00:40:07 +08:00
committed by GitHub
co-authored by ltd0924
parent c781db0f6c
commit 980d2936cd
15 changed files with 1557 additions and 12 deletions
@@ -466,11 +466,12 @@ class MultiLayerEagleDraftWorker(BaseDraftWorker):
draft_logits_output.topk_index,
)
else:
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
)
probs = torch.softmax(
draft_logits_output.next_token_logits[select_index], dim=-1
draft_logits_output.logits_output.next_token_logits[select_index],
dim=-1,
)
ret_topk_p, ret_topk_index = fast_topk(probs, self.topk, dim=-1)
if forward_batch.extend_seq_lens is not None: