File size: 5,866 Bytes
e2a9bea
 
 
 
 
 
 
8cfcd80
 
e2a9bea
 
 
 
 
8cfcd80
e2a9bea
 
8cfcd80
 
 
e2a9bea
 
 
 
 
 
 
 
 
 
 
 
 
8cfcd80
e2a9bea
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
import argparse
import markdown2
import os
import sys
import uvicorn

from pathlib import Path
from typing import Union

from fastapi import FastAPI, Depends
from fastapi.responses import HTMLResponse
from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
from pydantic import BaseModel, Field
from sse_starlette.sse import EventSourceResponse, ServerSentEvent

from messagers.message_composer import MessageComposer
from mocks.stream_chat_mocker import stream_chat_mock
from networks.message_streamer import MessageStreamer
from utils.logger import logger
from constants.models import AVAILABLE_MODELS_DICTS


class ChatAPIApp:
    def __init__(self):
        self.app = FastAPI(
            docs_url="/",
            title="HuggingFace LLM API",
            swagger_ui_parameters={"defaultModelsExpandDepth": -1},
            version="1.0",
        )
        self.setup_routes()

    def get_available_models(self):
        return {"object": "list", "data": AVAILABLE_MODELS_DICTS}

    def extract_api_key(
        credentials: HTTPAuthorizationCredentials = Depends(
            HTTPBearer(auto_error=False)
        ),
    ):
        api_key = None
        if credentials:
            api_key = credentials.credentials
        else:
            api_key = os.getenv("HF_TOKEN")

        if api_key:
            if api_key.startswith("hf_"):
                return api_key
            else:
                logger.warn(f"Invalid HF Token!")
        else:
            logger.warn("Not provide HF Token!")
        return None

    class ChatCompletionsPostItem(BaseModel):
        model: str = Field(
            default="mixtral-8x7b",
            description="(str) `mixtral-8x7b`",
        )
        messages: list = Field(
            default=[{"role": "user", "content": "Hello, who are you?"}],
            description="(list) Messages",
        )
        temperature: Union[float, None] = Field(
            default=0.5,
            description="(float) Temperature",
        )
        top_p: Union[float, None] = Field(
            default=0.95,
            description="(float) top p",
        )
        max_tokens: Union[int, None] = Field(
            default=-1,
            description="(int) Max tokens",
        )
        use_cache: bool = Field(
            default=False,
            description="(bool) Use cache",
        )
        stream: bool = Field(
            default=True,
            description="(bool) Stream",
        )

    def chat_completions(
        self, item: ChatCompletionsPostItem, api_key: str = Depends(extract_api_key)
    ):
        streamer = MessageStreamer(model=item.model)
        composer = MessageComposer(model=item.model)
        composer.merge(messages=item.messages)
        # streamer.chat = stream_chat_mock

        stream_response = streamer.chat_response(
            prompt=composer.merged_str,
            temperature=item.temperature,
            top_p=item.top_p,
            max_new_tokens=item.max_tokens,
            api_key=api_key,
            use_cache=item.use_cache,
        )
        if item.stream:
            event_source_response = EventSourceResponse(
                streamer.chat_return_generator(stream_response),
                media_type="text/event-stream",
                ping=2000,
                ping_message_factory=lambda: ServerSentEvent(**{"comment": ""}),
            )
            return event_source_response
        else:
            data_response = streamer.chat_return_dict(stream_response)
            return data_response

    def get_readme(self):
        readme_path = Path(__file__).parents[1] / "README.md"
        with open(readme_path, "r", encoding="utf-8") as rf:
            readme_str = rf.read()
        readme_html = markdown2.markdown(
            readme_str, extras=["table", "fenced-code-blocks", "highlightjs-lang"]
        )
        return readme_html

    def setup_routes(self):
        for prefix in ["", "/v1", "/api", "/api/v1"]:
            if prefix in ["/api/v1"]:
                include_in_schema = True
            else:
                include_in_schema = False

            self.app.get(
                prefix + "/models",
                summary="Get available models",
                include_in_schema=include_in_schema,
            )(self.get_available_models)

            self.app.post(
                prefix + "/chat/completions",
                summary="Chat completions in conversation session",
                include_in_schema=include_in_schema,
            )(self.chat_completions)
        self.app.get(
            "/readme",
            summary="README of HF LLM API",
            response_class=HTMLResponse,
            include_in_schema=False,
        )(self.get_readme)


class ArgParser(argparse.ArgumentParser):
    def __init__(self, *args, **kwargs):
        super(ArgParser, self).__init__(*args, **kwargs)

        self.add_argument(
            "-s",
            "--server",
            type=str,
            default="0.0.0.0",
            help="Server IP for HF LLM Chat API",
        )
        self.add_argument(
            "-p",
            "--port",
            type=int,
            default=23333,
            help="Server Port for HF LLM Chat API",
        )

        self.add_argument(
            "-d",
            "--dev",
            default=False,
            action="store_true",
            help="Run in dev mode",
        )

        self.args = self.parse_args(sys.argv[1:])


app = ChatAPIApp().app

if __name__ == "__main__":
    args = ArgParser().args
    if args.dev:
        uvicorn.run("__main__:app", host=args.server, port=args.port, reload=True)
    else:
        uvicorn.run("__main__:app", host=args.server, port=args.port, reload=False)

    # python -m apis.chat_api      # [Docker] on product mode
    # python -m apis.chat_api -d   # [Dev]    on develop mode