Support dynamic LoRA loading / unloading in engine/server API (#7446)

This commit is contained in:
Lifu Huang
2025-06-27 21:00:27 -07:00
committed by GitHub
parent cfe2edac38
commit 49538d111b
14 changed files with 949 additions and 31 deletions

View File

@@ -20,7 +20,7 @@ import copy
import uuid
from dataclasses import dataclass, field
from enum import Enum
from typing import TYPE_CHECKING, Any, Dict, List, Optional, Union
from typing import TYPE_CHECKING, Any, Dict, List, Optional, Set, Union
from sglang.srt.multimodal.mm_utils import has_valid_data
@@ -1002,3 +1002,27 @@ class RpcReqInput:
class RpcReqOutput:
success: bool
message: str
@dataclass
class LoadLoRAAdapterReqInput:
# The name of the lora module to newly loaded.
lora_name: str
# The path of loading.
lora_path: str
@dataclass
class UnloadLoRAAdapterReqInput:
# The name of lora module to unload.
lora_name: str
@dataclass
class LoRAUpdateResult:
success: bool
error_message: Optional[str] = None
loaded_adapters: Dict[str, str] = field(default_factory=dict)
LoadLoRAAdapterReqOutput = UnloadLoRAAdapterReqOutput = LoRAUpdateResult