From f7b1d8c5ab4d9f38d364490bcc31014bb8d790fe Mon Sep 17 00:00:00 2001 From: Qiaolin Yu Date: Mon, 20 Oct 2025 10:34:38 -0700 Subject: [PATCH] Fix acc len and gen throughput metrics when enabling overlap-spec (#11823) Co-authored-by: Liangsheng Yin --- .../sglang/srt/managers/scheduler_output_processor_mixin.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/python/sglang/srt/managers/scheduler_output_processor_mixin.py b/python/sglang/srt/managers/scheduler_output_processor_mixin.py index d1d78efb6..b238f6c25 100644 --- a/python/sglang/srt/managers/scheduler_output_processor_mixin.py +++ b/python/sglang/srt/managers/scheduler_output_processor_mixin.py @@ -208,7 +208,7 @@ class SchedulerOutputProcessorMixin: next_token_ids = result.next_token_ids.tolist() accept_lens = result.accept_lens.tolist() - result.num_accepted_tokens = sum(accept_lens) + result.num_accepted_tokens = sum(accept_lens) - len(batch.reqs) predict_tokens = [] stride = self.draft_worker.speculative_num_draft_tokens @@ -244,7 +244,7 @@ class SchedulerOutputProcessorMixin: accept_lens_list = result.accept_lens.tolist() self.num_generated_tokens += len(batch.reqs) - if not self.spec_algorithm.is_none(): + if not batch.spec_algorithm.is_none(): self.update_spec_metrics(batch.batch_size(), result.num_accepted_tokens) self.token_to_kv_pool_allocator.free_group_begin()