[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:
co-authored by
Yifan Qiao
Claude Opus 5.5
parent
08a65e287e
commit
c3de0b37e2
@@ -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
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user