[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
+24
View File
@@ -92,6 +92,7 @@ def test_get_kv_cache_spec_resolves_hisparse_block_size(
layers[name] = SimpleNamespace(
get_kv_cache_spec=lambda _, spec=specs[name]: spec,
get_attn_backend=lambda backend=backend: backend,
is_draft_layer=False,
)
monkeypatch.setattr(attn_utils, "get_layers_from_vllm_config", lambda *_: layers)
monkeypatch.setattr(
@@ -112,6 +113,29 @@ def test_get_kv_cache_spec_resolves_hisparse_block_size(
assert all(spec.block_size == block_size for spec in specs.values())
def test_get_kv_cache_spec_tags_draft_layers(monkeypatch):
spec = FullAttentionSpec(
block_size=16, num_kv_heads=1, head_size=128, dtype=torch.bfloat16
)
backend = SimpleNamespace(
customize_spec=AttentionBackend.customize_spec,
get_supported_kernel_block_sizes=lambda _=None: [16],
)
layers = {
name: SimpleNamespace(
get_kv_cache_spec=lambda _: spec,
get_attn_backend=lambda: backend,
is_draft_layer=name == "draft",
)
for name in ("target", "draft")
}
monkeypatch.setattr(attn_utils, "get_layers_from_vllm_config", lambda *_: layers)
config = SimpleNamespace(attention_config=SimpleNamespace(hisparse_config=None))
specs = attn_utils.get_kv_cache_spec(config)
assert specs["draft"].is_draft and not specs["target"].is_draft
class _FakeMetadataBuilder:
def __init__(self, support: AttentionCGSupport, varlen_bound: int | None = None):
self.support = support