mirror of
https://github.com/RYDE-WORK/Langchain-Chatchat.git
synced 2026-01-23 23:40:03 +08:00
b4c68ddd05
* 优化在线 API ，支持 completion 和 embedding，简化在线 API 开发方式 新功能 - 智谱AI、Minimax、千帆、千问 4 个在线模型支持 embeddings（不通过Fastchat，后续会单独提供相关api接口） - 在线模型自动检测传入参数，在传入非 messages 格式的 prompt 时，自动转换为 completion 形式，以支持 completion 接口 开发者： - 重构ApiModelWorker: - 所有在线 API 请求封装到 do_chat 方法：自动传入参数 ApiChatParams，简化参数与配置项的获取；自动处理与fastchat的接口 - 加强 API 请求错误处理，返回更有意义的信息 - 改用 qianfan sdk 重写 qianfan-api - 将所有在线模型的测试用例统一在一起，简化测试用例编写 * Delete requirements_langflow.txt
135 lines
4.9 KiB
Python
135 lines
4.9 KiB
Python
from fastchat.conversation import Conversation
|
||
from server.model_workers.base import *
|
||
from fastchat import conversation as conv
|
||
import sys
|
||
import json
|
||
from server.model_workers.base import ApiEmbeddingsParams
|
||
from server.utils import get_httpx_client
|
||
from typing import List, Dict
|
||
|
||
|
||
class MiniMaxWorker(ApiModelWorker):
|
||
DEFAULT_EMBED_MODEL = "embo-01"
|
||
|
||
def __init__(
|
||
self,
|
||
*,
|
||
model_names: List[str] = ["minimax-api"],
|
||
controller_addr: str = None,
|
||
worker_addr: str = None,
|
||
version: str = "abab5.5-chat",
|
||
**kwargs,
|
||
):
|
||
kwargs.update(model_names=model_names, controller_addr=controller_addr, worker_addr=worker_addr)
|
||
kwargs.setdefault("context_len", 16384)
|
||
super().__init__(**kwargs)
|
||
self.version = version
|
||
|
||
def validate_messages(self, messages: List[Dict]) -> List[Dict]:
|
||
role_maps = {
|
||
"user": self.user_role,
|
||
"assistant": self.ai_role,
|
||
"system": "system",
|
||
}
|
||
messages = [{"sender_type": role_maps[x["role"]], "text": x["content"]} for x in messages]
|
||
return messages
|
||
|
||
def do_chat(self, params: ApiChatParams) -> Dict:
|
||
# 按照官网推荐，直接调用abab 5.5模型
|
||
# TODO: 支持指定回复要求，支持指定用户名称、AI名称
|
||
params.load_config(self.model_names[0])
|
||
|
||
url = 'https://api.minimax.chat/v1/text/chatcompletion{pro}?GroupId={group_id}'
|
||
pro = "_pro" if params.is_pro else ""
|
||
headers = {
|
||
"Authorization": f"Bearer {params.api_key}",
|
||
"Content-Type": "application/json",
|
||
}
|
||
messages = self.validate_messages(params.messages)
|
||
data = {
|
||
"model": params.version,
|
||
"stream": True,
|
||
"mask_sensitive_info": True,
|
||
"messages": messages,
|
||
"temperature": params.temperature,
|
||
"top_p": params.top_p,
|
||
"tokens_to_generate": params.max_tokens or 1024,
|
||
# TODO: 以下参数为minimax特有，传入空值会出错。
|
||
# "prompt": params.system_message or self.conv.system_message,
|
||
# "bot_setting": [],
|
||
# "role_meta": params.role_meta,
|
||
}
|
||
|
||
with get_httpx_client() as client:
|
||
response = client.stream("POST",
|
||
url.format(pro=pro, group_id=params.group_id),
|
||
headers=headers,
|
||
json=data)
|
||
with response as r:
|
||
text = ""
|
||
for e in r.iter_text():
|
||
if not e.startswith("data: "): # 真是优秀的返回
|
||
yield {"error_code": 500, "text": f"minimax返回错误的结果：{e}"}
|
||
continue
|
||
|
||
data = json.loads(e[6:])
|
||
if data.get("usage"):
|
||
break
|
||
|
||
if choices := data.get("choices"):
|
||
if chunk := choices[0].get("delta", ""):
|
||
text += chunk
|
||
yield {"error_code": 0, "text": text}
|
||
|
||
def do_embeddings(self, params: ApiEmbeddingsParams) -> Dict:
|
||
params.load_config(self.model_names[0])
|
||
url = f"https://api.minimax.chat/v1/embeddings?GroupId={params.group_id}"
|
||
|
||
headers = {
|
||
"Authorization": f"Bearer {params.api_key}",
|
||
"Content-Type": "application/json",
|
||
}
|
||
|
||
data = {
|
||
"model": params.embed_model or self.DEFAULT_EMBED_MODEL,
|
||
"texts": params.texts,
|
||
"type": "query" if params.to_query else "db",
|
||
}
|
||
|
||
with get_httpx_client() as client:
|
||
r = client.post(url, headers=headers, json=data).json()
|
||
if embeddings := r.get("vectors"):
|
||
return {"code": 200, "embeddings": embeddings}
|
||
elif error := r.get("base_resp"):
|
||
return {"code": error["status_code"], "msg": error["status_msg"]}
|
||
|
||
def get_embeddings(self, params):
|
||
# TODO: 支持embeddings
|
||
print("embedding")
|
||
print(params)
|
||
|
||
def make_conv_template(self, conv_template: str = None, model_path: str = None) -> Conversation:
|
||
# TODO: 确认模板是否需要修改
|
||
return conv.Conversation(
|
||
name=self.model_names[0],
|
||
system_message="你是MiniMax自主研发的大型语言模型，回答问题简洁有条理。",
|
||
messages=[],
|
||
roles=["USER", "BOT"],
|
||
sep="\n### ",
|
||
stop_str="###",
|
||
)
|
||
|
||
|
||
if __name__ == "__main__":
|
||
import uvicorn
|
||
from server.utils import MakeFastAPIOffline
|
||
from fastchat.serve.model_worker import app
|
||
|
||
worker = MiniMaxWorker(
|
||
controller_addr="http://127.0.0.1:20001",
|
||
worker_addr="http://127.0.0.1:21002",
|
||
)
|
||
sys.modules["fastchat.serve.model_worker"].worker = worker
|
||
MakeFastAPIOffline(app)
|
||
uvicorn.run(app, port=21002)
|