From 27c3190215dfefc6ec5d5d8d78874ebb76a67bc9 Mon Sep 17 00:00:00 2001 From: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com> Date: Sat, 16 May 2026 09:18:00 +0800 Subject: [PATCH] Remove fields that are never used in spec/engine/disagg (#25434) --- python/sglang/srt/disaggregation/kv_events.py | 4 ---- python/sglang/srt/entrypoints/engine.py | 5 ----- python/sglang/srt/speculative/adaptive_spec_params.py | 2 -- python/sglang/srt/speculative/cpp_ngram/ngram_corpus.py | 1 - python/sglang/srt/speculative/eagle_worker_v2.py | 1 - 5 files changed, 13 deletions(-) diff --git a/python/sglang/srt/disaggregation/kv_events.py b/python/sglang/srt/disaggregation/kv_events.py index 0a5549bb1..ef7dda684 100644 --- a/python/sglang/srt/disaggregation/kv_events.py +++ b/python/sglang/srt/disaggregation/kv_events.py @@ -119,9 +119,6 @@ class EventPublisher(ABC): - This allows consumers to distinguish events from different DP ranks """ - def __init__(self, attn_dp_rank: int = 0): - self._attn_dp_rank = attn_dp_rank - @abstractmethod def publish(self, events: EventBatch) -> None: """Emit events in order. @@ -183,7 +180,6 @@ class ZmqEventPublisher(EventPublisher): topic: str = "", ) -> None: # Storage - super().__init__(attn_dp_rank) self._event_queue = Queue[Optional[EventBatch]](maxsize=max_queue_size) self._buffer = deque[tuple[int, bytes]](maxlen=buffer_steps) diff --git a/python/sglang/srt/entrypoints/engine.py b/python/sglang/srt/entrypoints/engine.py index 1dde8bed8..ee84ff105 100644 --- a/python/sglang/srt/entrypoints/engine.py +++ b/python/sglang/srt/entrypoints/engine.py @@ -245,11 +245,6 @@ class Engine(EngineScoreMixin, EngineBase): if tokenizer_manager is not None: tokenizer_manager._subprocess_watchdog = subprocess_watchdog self.port_args = port_args - # Access transfer engine info if bootstrap server is started. - if scheduler_init_result.engine_info_bootstrap_server is not None: - self.remote_instance_transfer_engine_info = ( - scheduler_init_result.engine_info_bootstrap_server.transfer_engine_info - ) # Initialize ZMQ sockets context = zmq.Context(2) diff --git a/python/sglang/srt/speculative/adaptive_spec_params.py b/python/sglang/srt/speculative/adaptive_spec_params.py index 37bae4e4d..d92795773 100644 --- a/python/sglang/srt/speculative/adaptive_spec_params.py +++ b/python/sglang/srt/speculative/adaptive_spec_params.py @@ -112,8 +112,6 @@ class AdaptiveSpeculativeParams: len(self.candidate_steps) >= 2 ), "candidate_steps must have at least 2 distinct values" - self.min_steps = self.candidate_steps[0] - self.max_steps = self.candidate_steps[-1] self.ema_alpha = cfg.get("ema_alpha", 0.2) self.update_interval = cfg.get("update_interval", 5) self.warmup_batches = cfg.get("warmup_batches", 10) diff --git a/python/sglang/srt/speculative/cpp_ngram/ngram_corpus.py b/python/sglang/srt/speculative/cpp_ngram/ngram_corpus.py index 4ffc299fd..e5a4b5656 100644 --- a/python/sglang/srt/speculative/cpp_ngram/ngram_corpus.py +++ b/python/sglang/srt/speculative/cpp_ngram/ngram_corpus.py @@ -34,7 +34,6 @@ class NgramCorpus: external_sam_budget=external_sam_budget, external_corpus_max_tokens=external_corpus_max_tokens, ) - self.default_mask = np.ones((1, 1), dtype=np.int64) self.draft_token_num = draft_token_num self.external_corpus_max_tokens = external_corpus_max_tokens self._req_id_to_state_id: Dict[str, int] = {} diff --git a/python/sglang/srt/speculative/eagle_worker_v2.py b/python/sglang/srt/speculative/eagle_worker_v2.py index e845fde3f..ef23afd74 100644 --- a/python/sglang/srt/speculative/eagle_worker_v2.py +++ b/python/sglang/srt/speculative/eagle_worker_v2.py @@ -240,7 +240,6 @@ class EagleDraftWorker(BaseDraftWorker): def init_attention_backend(self): # Create multi-step attn backends and cuda graph runners - self.has_prefill_wrapper_verify = False self.draft_extend_attn_backend = None draft_backend_factory = DraftBackendFactory(