gpt4free/g4f/api/__init__.py

201 lines
8.0 KiB
Python
Raw Normal View History

import ast
import logging
import time
2023-10-12 04:35:11 +03:00
import json
import random
import string
2023-11-02 04:27:35 +03:00
import uvicorn
import nest_asyncio
from fastapi import FastAPI, Response, Request
from fastapi.responses import StreamingResponse
from typing import List, Union, Any, Dict, AnyStr
#from ._tokenizer import tokenize
2023-11-05 00:16:09 +03:00
import g4f
from .. import debug
debug.logging = True
2023-11-02 04:27:35 +03:00
2023-11-05 00:16:09 +03:00
class Api:
def __init__(self, engine: g4f, debug: bool = True, sentry: bool = False,
list_ignored_providers: List[str] = None) -> None:
2023-11-05 00:16:09 +03:00
self.engine = engine
self.debug = debug
self.sentry = sentry
self.list_ignored_providers = list_ignored_providers
self.app = FastAPI()
nest_asyncio.apply()
JSONObject = Dict[AnyStr, Any]
JSONArray = List[Any]
JSONStructure = Union[JSONArray, JSONObject]
@self.app.get("/")
async def read_root():
return Response(content=json.dumps({"info": "g4f API"}, indent=4), media_type="application/json")
@self.app.get("/v1")
async def read_root_v1():
return Response(content=json.dumps({"info": "Go to /v1/chat/completions or /v1/models."}, indent=4), media_type="application/json")
@self.app.get("/v1/models")
async def models():
model_list = []
for model in g4f.Model.__all__():
model_info = (g4f.ModelUtils.convert[model])
model_list.append({
2023-11-05 00:16:09 +03:00
'id': model,
'object': 'model',
'created': 0,
'owned_by': model_info.base_provider}
)
2023-11-05 00:16:09 +03:00
return Response(content=json.dumps({
'object': 'list',
'data': model_list}, indent=4), media_type="application/json")
@self.app.get("/v1/models/{model_name}")
async def model_info(model_name: str):
try:
model_info = (g4f.ModelUtils.convert[model_name])
return Response(content=json.dumps({
'id': model_name,
'object': 'model',
'created': 0,
'owned_by': model_info.base_provider
}, indent=4), media_type="application/json")
except:
return Response(content=json.dumps({"error": "The model does not exist."}, indent=4), media_type="application/json")
@self.app.post("/v1/chat/completions")
async def chat_completions(request: Request, item: JSONStructure = None):
item_data = {
'model': 'gpt-3.5-turbo',
'stream': False,
}
# item contains byte keys, and dict.get suppresses error
item_data.update({
key.decode('utf-8') if isinstance(key, bytes) else key: str(value)
for key, value in (item or {}).items()
})
# messages is str, need dict
if isinstance(item_data.get('messages'), str):
item_data['messages'] = ast.literal_eval(item_data.get('messages'))
2023-11-05 00:16:09 +03:00
model = item_data.get('model')
stream = True if item_data.get("stream") == "True" else False
2023-11-05 00:16:09 +03:00
messages = item_data.get('messages')
2023-12-23 22:50:56 +03:00
provider = item_data.get('provider', '').replace('g4f.Provider.', '')
2023-12-14 05:16:35 +03:00
provider = provider if provider and provider != "Auto" else None
temperature = item_data.get('temperature')
2023-11-05 00:16:09 +03:00
try:
2023-12-23 22:50:56 +03:00
response = g4f.ChatCompletion.create(
model=model,
stream=stream,
messages=messages,
temperature = temperature,
2023-12-23 22:50:56 +03:00
provider = provider,
ignored=self.list_ignored_providers
)
except Exception as e:
logging.exception(e)
content = json.dumps({
"error": {"message": f"An error occurred while generating the response:\n{e}"},
"model": model,
"provider": g4f.get_last_provider(True)
})
return Response(content=content, status_code=500, media_type="application/json")
2023-11-05 00:16:09 +03:00
completion_id = ''.join(random.choices(string.ascii_letters + string.digits, k=28))
completion_timestamp = int(time.time())
if not stream:
#prompt_tokens, _ = tokenize(''.join([message['content'] for message in messages]))
#completion_tokens, _ = tokenize(response)
2023-11-05 00:16:09 +03:00
json_data = {
2023-11-02 04:27:35 +03:00
'id': f'chatcmpl-{completion_id}',
2023-11-05 00:16:09 +03:00
'object': 'chat.completion',
2023-11-02 04:27:35 +03:00
'created': completion_timestamp,
'model': model,
'provider': g4f.get_last_provider(True),
2023-11-02 04:27:35 +03:00
'choices': [
{
'index': 0,
2023-11-05 00:16:09 +03:00
'message': {
'role': 'assistant',
'content': response,
2023-11-02 04:27:35 +03:00
},
2023-11-05 00:16:09 +03:00
'finish_reason': 'stop',
2023-11-02 04:27:35 +03:00
}
],
2023-11-05 00:16:09 +03:00
'usage': {
'prompt_tokens': 0, #prompt_tokens,
'completion_tokens': 0, #completion_tokens,
'total_tokens': 0, #prompt_tokens + completion_tokens,
2023-11-05 00:16:09 +03:00
},
2023-11-02 04:27:35 +03:00
}
2023-10-27 12:33:47 +03:00
2023-11-05 00:16:09 +03:00
return Response(content=json.dumps(json_data, indent=4), media_type="application/json")
def streaming():
try:
for chunk in response:
completion_data = {
'id': f'chatcmpl-{completion_id}',
'object': 'chat.completion.chunk',
'created': completion_timestamp,
'model': model,
'provider': g4f.get_last_provider(True),
2023-11-05 00:16:09 +03:00
'choices': [
{
'index': 0,
'delta': {
'role': 'assistant',
2023-11-05 00:16:09 +03:00
'content': chunk,
},
'finish_reason': None,
}
],
}
yield f'data: {json.dumps(completion_data)}\n\n'
2023-11-05 00:16:09 +03:00
time.sleep(0.03)
end_completion_data = {
'id': f'chatcmpl-{completion_id}',
'object': 'chat.completion.chunk',
'created': completion_timestamp,
'model': model,
'provider': g4f.get_last_provider(True),
2023-11-05 00:16:09 +03:00
'choices': [
{
'index': 0,
'delta': {},
'finish_reason': 'stop',
}
],
2023-10-12 04:35:11 +03:00
}
yield f'data: {json.dumps(end_completion_data)}\n\n'
2023-11-05 00:16:09 +03:00
except GeneratorExit:
pass
2023-12-23 22:50:56 +03:00
except Exception as e:
logging.exception(e)
content = json.dumps({
"error": {"message": f"An error occurred while generating the response:\n{e}"},
"model": model,
"provider": g4f.get_last_provider(True),
})
yield f'data: {content}'
return StreamingResponse(streaming(), media_type="text/event-stream")
2023-11-05 00:16:09 +03:00
@self.app.post("/v1/completions")
async def completions():
return Response(content=json.dumps({'info': 'Not working yet.'}, indent=4), media_type="application/json")
2023-11-02 04:27:35 +03:00
2023-11-05 00:26:16 +03:00
def run(self, ip):
2023-11-05 00:16:09 +03:00
split_ip = ip.split(":")
2023-11-05 00:27:25 +03:00
uvicorn.run(app=self.app, host=split_ip[0], port=int(split_ip[1]), use_colors=False)