[Fast Start] Cache the MTP draft model in a separate daemon group (#57312)

Signed-off-by: liusy58 <mg21330037@smail.nju.edu.cn>
Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: Isotr0py <Isotr0py@outlook.com>
This commit is contained in:
siyu
2026-09-22 03:13:38 +00:00
committed by GitHub
co-authored by Isotr0py
parent f07e227ee9
commit c723a831a8
16 changed files with 417 additions and 140 deletions
@@ -12,7 +12,7 @@ from unittest.mock import patch
import pytest
from vllm.config import LoadConfig
from vllm.config import LoadConfig, SpeculativeConfig
from vllm.v1.worker.gpu.spec_decode.eagle.utils import load_eagle_model
@@ -37,6 +37,9 @@ class _SpeculativeConfig:
moe_backend: str | None = None
kv_cache_dtype: str | None = None
draft_model_config: object = None
draft_load_config: object = None
apply_draft_overrides = SpeculativeConfig.apply_draft_overrides
@dataclass
@@ -14,7 +14,7 @@ from unittest.mock import patch
import pytest
from vllm.config import LoadConfig
from vllm.config import LoadConfig, SpeculativeConfig
from vllm.v1.worker.gpu.spec_decode.eagle.utils import load_eagle_model
@@ -34,6 +34,9 @@ class _SpeculativeConfig:
moe_backend: str | None = None
kv_cache_dtype: str | None = None
draft_model_config: object = None
draft_load_config: object = None
apply_draft_overrides = SpeculativeConfig.apply_draft_overrides
@dataclass