From b6d8e8afd985f5711eee68e343d2ce908d166488 Mon Sep 17 00:00:00 2001 From: Cheese Cake Date: Wed, 7 Oct 2026 03:32:04 +0530 Subject: [PATCH] [Bugfix][MRV2][Spec Decode] Accept num_speculative_tokens in NgramGPUSpeculator.propose (#60300) Signed-off-by: cheesecake Co-authored-by: Claude Opus 5.5 --- vllm/v1/worker/gpu/spec_decode/ngram/speculator.py | 1 + 1 file changed, 1 insertion(+) diff --git a/vllm/v1/worker/gpu/spec_decode/ngram/speculator.py b/vllm/v1/worker/gpu/spec_decode/ngram/speculator.py index 404f576274..7a61e3581a 100644 --- a/vllm/v1/worker/gpu/spec_decode/ngram/speculator.py +++ b/vllm/v1/worker/gpu/spec_decode/ngram/speculator.py @@ -231,6 +231,7 @@ class NgramGPUSpeculator(BaseSpeculator): skip_attn_for_dummy_run: bool = False, mm_inputs: tuple[list[torch.Tensor], torch.Tensor] | None = None, is_profile: bool = False, + num_speculative_tokens: int | None = None, ) -> torch.Tensor: num_reqs = input_batch.num_reqs if dummy_run: