[Core] Configure custom encoder cache managers from VllmConfig (#51251)

Signed-off-by: hotTea <958436561@qq.com>
Signed-off-by: HMCCMH <chenminghaoscu@163.com>
Co-authored-by: HMCCMH <chenminghaoscu@163.com>
This commit is contained in:
hotTea
2026-08-13 16:08:01 +08:00
committed by GitHub
co-authored by HMCCMH
parent 399f97424e
commit 7bbbf7c8e5
4 changed files with 36 additions and 8 deletions
+10
View File
@@ -41,6 +41,7 @@ from vllm.config import (
DeviceConfig,
DiffusionConfig,
ECTransferConfig,
EncoderCacheManagerConfig,
EPLBConfig,
FaultToleranceConfig,
KernelConfig,
@@ -686,6 +687,9 @@ class EngineArgs:
kv_events_config: KVEventsConfig | None = None
ec_transfer_config: ECTransferConfig | None = None
ec_manager_config: EncoderCacheManagerConfig = get_field(
VllmConfig, "ec_manager_config"
)
reasoning_config: ReasoningConfig = get_field(VllmConfig, "reasoning_config")
generation_config: str = ModelConfig.generation_config
@@ -766,6 +770,8 @@ class EngineArgs:
self.mamba_config = MambaConfig(**self.mamba_config)
if isinstance(self.kernel_config, dict):
self.kernel_config = KernelConfig(**self.kernel_config)
if isinstance(self.ec_manager_config, dict):
self.ec_manager_config = EncoderCacheManagerConfig(**self.ec_manager_config)
if isinstance(self.eplb_config, dict):
self.eplb_config = EPLBConfig(**self.eplb_config)
if isinstance(self.weight_transfer_config, dict):
@@ -1625,6 +1631,9 @@ class EngineArgs:
vllm_group.add_argument(
"--ec-transfer-config", **vllm_kwargs["ec_transfer_config"]
)
vllm_group.add_argument(
"--ec-manager-config", **vllm_kwargs["ec_manager_config"]
)
vllm_group.add_argument(
"--compilation-config", "-cc", **vllm_kwargs["compilation_config"]
)
@@ -2506,6 +2515,7 @@ class EngineArgs:
kv_transfer_config=self.kv_transfer_config,
kv_events_config=self.kv_events_config,
ec_transfer_config=self.ec_transfer_config,
ec_manager_config=self.ec_manager_config,
reasoning_config=self.reasoning_config,
profiler_config=self.profiler_config,
additional_config=self.additional_config,