mirror of
https://github.com/csunny/DB-GPT.git
synced 2026-07-18 02:45:26 +00:00
53 lines
1.2 KiB
Python
53 lines
1.2 KiB
Python
from dataclasses import dataclass
|
|
from typing import Dict, List
|
|
|
|
from pilot.model.base import WorkerApplyType
|
|
from pilot.model.parameter import WorkerType
|
|
from pilot.scene.base_message import ModelMessage
|
|
from pydantic import BaseModel
|
|
|
|
WORKER_MANAGER_SERVICE_TYPE = "service"
|
|
WORKER_MANAGER_SERVICE_NAME = "WorkerManager"
|
|
|
|
|
|
class PromptRequest(BaseModel):
|
|
messages: List[ModelMessage]
|
|
model: str
|
|
prompt: str = None
|
|
temperature: float = None
|
|
max_new_tokens: int = None
|
|
stop: str = None
|
|
stop_token_ids: List[int] = []
|
|
context_len: int = None
|
|
echo: bool = True
|
|
span_id: str = None
|
|
metrics: bool = False
|
|
"""Whether to return metrics of inference"""
|
|
|
|
|
|
class EmbeddingsRequest(BaseModel):
|
|
model: str
|
|
input: List[str]
|
|
span_id: str = None
|
|
|
|
|
|
class WorkerApplyRequest(BaseModel):
|
|
model: str
|
|
apply_type: WorkerApplyType
|
|
worker_type: WorkerType = WorkerType.LLM
|
|
params: Dict = None
|
|
apply_user: str = None
|
|
|
|
|
|
class WorkerParameterRequest(BaseModel):
|
|
model: str
|
|
worker_type: WorkerType = WorkerType.LLM
|
|
|
|
|
|
class WorkerStartupRequest(BaseModel):
|
|
host: str
|
|
port: int
|
|
model: str
|
|
worker_type: WorkerType
|
|
params: Dict
|