File size: 4,198 Bytes
a16695d
 
 
 
 
 
 
 
 
 
0edea2e
 
 
e820e51
a16695d
37be019
 
a16695d
 
0edea2e
 
37be019
a16695d
acbd8bf
cec776f
0edea2e
e820e51
a16695d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37be019
a16695d
 
 
 
 
 
cec776f
 
37be019
 
a16695d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37be019
a16695d
 
 
 
 
 
 
 
 
 
37be019
 
 
 
 
a16695d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
title: HF LLM API
emoji: ☯️
colorFrom: gray
colorTo: gray
sdk: docker
app_port: 23333
---

## HF-LLM-API

![](https://img.shields.io/github/v/release/hansimov/hf-llm-api?label=HF-LLM-API&color=blue&cacheSeconds=60)

Huggingface LLM Inference API in OpenAI message format.

Project link: https://github.com/Hansimov/hf-llm-api

## Features

- Available Models (2024/04/07):
  - `mistral-7b`, `mixtral-8x7b`, `nous-mixtral-8x7b`, `gemma-7b`, `gpt-3.5.turbo`
  - Adaptive prompt templates for different models
- Support OpenAI API format
  - Enable api endpoint via official `openai-python` package
- Support both stream and no-stream response
- Support API Key via both HTTP auth header and env variable
- Docker deployment

## Run API service

### Run in Command Line

**Install dependencies:**

```bash
# pipreqs . --force --mode no-pin
pip install -r requirements.txt
```

**Run API:**

```bash
python -m apis.chat_api
```

## Run via Docker

**Docker build:**

```bash
sudo docker build -t hf-llm-api:1.0 . --build-arg http_proxy=$http_proxy --build-arg https_proxy=$https_proxy
```

**Docker run:**

```bash
# no proxy
sudo docker run -p 23333:23333 hf-llm-api:1.0

# with proxy
sudo docker run -p 23333:23333 --env http_proxy="http://<server>:<port>" hf-llm-api:1.0
```

## API Usage

### Using `openai-python`

See: [`examples/chat_with_openai.py`](https://github.com/Hansimov/hf-llm-api/blob/main/examples/chat_with_openai.py)

```py
from openai import OpenAI

# If runnning this service with proxy, you might need to unset `http(s)_proxy`.
base_url = "http://127.0.0.1:23333"
# Your own HF_TOKEN
api_key = "hf_xxxxxxxxxxxxxxxx"
# use below as non-auth user
# api_key = "sk-xxx"

client = OpenAI(base_url=base_url, api_key=api_key)
response = client.chat.completions.create(
    model="mixtral-8x7b",
    messages=[
        {
            "role": "user",
            "content": "what is your model",
        }
    ],
    stream=True,
)

for chunk in response:
    if chunk.choices[0].delta.content is not None:
        print(chunk.choices[0].delta.content, end="", flush=True)
    elif chunk.choices[0].finish_reason == "stop":
        print()
    else:
        pass
```

### Using post requests

See: [`examples/chat_with_post.py`](https://github.com/Hansimov/hf-llm-api/blob/main/examples/chat_with_post.py)


```py
import ast
import httpx
import json
import re

# If runnning this service with proxy, you might need to unset `http(s)_proxy`.
chat_api = "http://127.0.0.1:23333"
# Your own HF_TOKEN
api_key = "hf_xxxxxxxxxxxxxxxx"
# use below as non-auth user
# api_key = "sk-xxx"

requests_headers = {}
requests_payload = {
    "model": "mixtral-8x7b",
    "messages": [
        {
            "role": "user",
            "content": "what is your model",
        }
    ],
    "stream": True,
}

with httpx.stream(
    "POST",
    chat_api + "/chat/completions",
    headers=requests_headers,
    json=requests_payload,
    timeout=httpx.Timeout(connect=20, read=60, write=20, pool=None),
) as response:
    # https://docs.aiohttp.org/en/stable/streams.html
    # https://github.com/openai/openai-cookbook/blob/main/examples/How_to_stream_completions.ipynb
    response_content = ""
    for line in response.iter_lines():
        remove_patterns = [r"^\s*data:\s*", r"^\s*\[DONE\]\s*"]
        for pattern in remove_patterns:
            line = re.sub(pattern, "", line).strip()

        if line:
            try:
                line_data = json.loads(line)
            except Exception as e:
                try:
                    line_data = ast.literal_eval(line)
                except:
                    print(f"Error: {line}")
                    raise e
            # print(f"line: {line_data}")
            delta_data = line_data["choices"][0]["delta"]
            finish_reason = line_data["choices"][0]["finish_reason"]
            if "role" in delta_data:
                role = delta_data["role"]
            if "content" in delta_data:
                delta_content = delta_data["content"]
                response_content += delta_content
                print(delta_content, end="", flush=True)
            if finish_reason == "stop":
                print()

```