[Core][Spec Decode] Replace EAGLE group heuristics with explicit draft-layer tags (#61455)

Signed-off-by: Dakai An <dakaian108@gmail.com>
Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai>
Co-authored-by: Yifan Qiao <yifanqiao@inferact.ai>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Dakai An
2026-10-17 08:22:33 +00:00
committed by GitHub
co-authored by Yifan Qiao Claude Opus 5.5
parent 08a65e287e
commit c3de0b37e2
10 changed files with 160 additions and 354 deletions
+4 -1
View File
@@ -1330,10 +1330,13 @@ class SpecDecodeBaseProposer:
self.vllm_config,
AttentionLayerBase, # type: ignore[type-abstract]
)
draft_attn_layer_names = set(all_attn_layers) - target_attn_layer_names
for name in draft_attn_layer_names:
all_attn_layers[name].is_draft_layer = True
# Filter to only layers that have KV cache specs.
self._draft_attn_layer_names = {
name
for name in (set(all_attn_layers.keys()) - target_attn_layer_names)
for name in draft_attn_layer_names
if all_attn_layers[name].get_kv_cache_spec(self.vllm_config) is not None
}