From a77449f86d56f7c4b3420d4392e123286b1179bd Mon Sep 17 00:00:00 2001 From: Qiaolin Yu Date: Mon, 25 May 2026 02:06:48 -0700 Subject: [PATCH] [perf][spec decoding] Skip full-vocab softmax in EAGLE draft when topk == 1 (#26235) --- .../eagle_draft_extend_cuda_graph_runner.py | 10 +++++++-- .../sglang/srt/speculative/eagle_worker_v2.py | 22 +++++++++++++++---- 2 files changed, 26 insertions(+), 6 deletions(-) diff --git a/python/sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py b/python/sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py index 9df3742fc..e58842278 100644 --- a/python/sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py +++ b/python/sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py @@ -401,8 +401,14 @@ class EAGLEDraftExtendCudaGraphRunner: forward_batch.positions, forward_batch, ) - probs = torch.softmax(ret.next_token_logits, dim=-1) - ret.topk_p, ret.topk_index = fast_topk(probs, self.topk, dim=-1) + if self.topk == 1: + ret.topk_index = torch.argmax( + ret.next_token_logits, dim=-1, keepdim=True + ) + ret.topk_p = torch.ones_like(ret.topk_index, dtype=torch.float32) + else: + probs = torch.softmax(ret.next_token_logits, dim=-1) + ret.topk_p, ret.topk_index = fast_topk(probs, self.topk, dim=-1) forward_batch.out_cache_loc = output_cache_loc_backup forward_batch.spec_info.hidden_states = hidden_states_backup diff --git a/python/sglang/srt/speculative/eagle_worker_v2.py b/python/sglang/srt/speculative/eagle_worker_v2.py index eb5c40a77..c864dc12d 100644 --- a/python/sglang/srt/speculative/eagle_worker_v2.py +++ b/python/sglang/srt/speculative/eagle_worker_v2.py @@ -483,8 +483,16 @@ class EagleDraftWorker(BaseDraftWorker): forward_batch, skip_attn_backend_init=True ).logits_output maybe_detect_nan(logits_output.next_token_logits, f"draft_forward step {i}") - probs = torch.softmax(logits_output.next_token_logits, dim=-1) - topk_p, topk_index = fast_topk(probs, self.topk, dim=-1) + if self.topk == 1: + # topk=1 → degenerate single-path tree; `topk_p` is unused + # downstream, so skip softmax and just argmax over logits. + topk_index = torch.argmax( + logits_output.next_token_logits, dim=-1, keepdim=True + ) + topk_p = torch.ones_like(topk_index, dtype=torch.float32) + else: + probs = torch.softmax(logits_output.next_token_logits, dim=-1) + topk_p, topk_index = fast_topk(probs, self.topk, dim=-1) maybe_detect_oob( topk_index, 0, @@ -651,8 +659,14 @@ class EagleDraftWorker(BaseDraftWorker): draft_logits_output.hidden_states = draft_logits_output.hidden_states[ select_index ] - probs = torch.softmax(draft_logits_output.next_token_logits, dim=-1) - ret_topk_p, ret_topk_index = fast_topk(probs, self.topk, dim=-1) + if self.topk == 1: + ret_topk_index = torch.argmax( + draft_logits_output.next_token_logits, dim=-1, keepdim=True + ) + ret_topk_p = torch.ones_like(ret_topk_index, dtype=torch.float32) + else: + probs = torch.softmax(draft_logits_output.next_token_logits, dim=-1) + ret_topk_p, ret_topk_index = fast_topk(probs, self.topk, dim=-1) ret_hidden_states = draft_logits_output.hidden_states # Construct the return values