[Speculative Decoding] Support MTP for GLM-4.5-Air (#6047)

* glm mtp * add spec neox partial rope
2026-04-23 08:21:53 +08:00 · 2026-01-16 14:35:24 +08:00
parent b2a2e11551
commit bda38aa519
9 changed files with 627 additions and 31 deletions
@@ -465,6 +465,8 @@ def post_process_specualate(
            step_idx=share_inputs["step_idx"],
            limit_think_status=share_inputs["limit_think_status"],
            accept_num=share_inputs["accept_num"],
+            stop_flags=share_inputs["stop_flags"],
+            eos_token_ids=share_inputs["eos_token_id"],
            think_end_id=think_end_id,
            line_break_id=line_break_id,
        )