[Bugfix][MRV2][Spec Decode] Honor EAGLE3 single-state draft configuration (#61563)

Signed-off-by: Sandeep Agrawal <sandeepa@nvidia.com>
Co-authored-by: Kevin H. Luu <khluu000@gmail.com>
This commit is contained in:
Sandeep Agrawal
2026-10-14 12:09:06 -07:00
committed by GitHub
co-authored by Kevin H. Luu
parent 06265beb25
commit 21e207efa1
3 changed files with 39 additions and 8 deletions
+4 -8
View File
@@ -156,6 +156,7 @@ from vllm.v1.worker.gpu.spec_decode.adaptive_verification import (
resolve_adaptive_cudagraph_mode,
)
from vllm.v1.worker.gpu.spec_decode.eagle.eagle3_utils import (
requires_aux_hidden_states,
set_eagle3_aux_hidden_state_layers,
verify_supports_aux_hidden_states_over_pp,
)
@@ -297,14 +298,9 @@ class GPUModelRunner(LoRAModelRunnerMixin):
self.vllm_config, self.device, self.req_states
)
if self.speculative_config.method in (
"eagle3",
"dflash",
"dspark",
"extract_hidden_states",
):
# Drafting may require auxiliary hidden states from target model outputs
self.use_aux_hidden_state_outputs = True
self.use_aux_hidden_state_outputs = requires_aux_hidden_states(
self.speculative_config
)
# Draft tokens propagation - for spec-dec + struct outputs.
self.draft_tokens_handler = DraftTokensHandler(self.device)