Spaces:
Running
Running
init
Browse files- Dockerfile +11 -0
- README.md +11 -11
- main.py +76 -0
- requirements.txt +3 -0
Dockerfile
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
FROM python:3.9
|
| 2 |
+
|
| 3 |
+
WORKDIR /code
|
| 4 |
+
|
| 5 |
+
COPY ./requirements.txt /code/requirements.txt
|
| 6 |
+
|
| 7 |
+
RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
|
| 8 |
+
|
| 9 |
+
COPY . .
|
| 10 |
+
|
| 11 |
+
CMD ["gunicorn", "-b", "0.0.0.0:7860", "main:app"]
|
README.md
CHANGED
|
@@ -1,11 +1,11 @@
|
|
| 1 |
-
---
|
| 2 |
-
title:
|
| 3 |
-
emoji:
|
| 4 |
-
colorFrom:
|
| 5 |
-
colorTo:
|
| 6 |
-
sdk: docker
|
| 7 |
-
pinned: false
|
| 8 |
-
license: apache-2.0
|
| 9 |
-
---
|
| 10 |
-
|
| 11 |
-
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
|
| 1 |
+
---
|
| 2 |
+
title: Huggingface Chat API Adapter
|
| 3 |
+
emoji: 💻
|
| 4 |
+
colorFrom: yellow
|
| 5 |
+
colorTo: indigo
|
| 6 |
+
sdk: docker
|
| 7 |
+
pinned: false
|
| 8 |
+
license: apache-2.0
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
main.py
ADDED
|
@@ -0,0 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import random
|
| 2 |
+
import requests
|
| 3 |
+
from flask import Flask, request, Response, stream_with_context, render_template_string
|
| 4 |
+
|
| 5 |
+
app = Flask(__name__)
|
| 6 |
+
|
| 7 |
+
@app.route('/', methods=['GET'])
|
| 8 |
+
def index():
|
| 9 |
+
template = '''
|
| 10 |
+
<html>
|
| 11 |
+
<head>
|
| 12 |
+
<title>Huggingface Chat API Adapter</title>
|
| 13 |
+
</head>
|
| 14 |
+
<body>
|
| 15 |
+
<h1>Huggingface Chat API Adapter</h1>
|
| 16 |
+
|
| 17 |
+
[Introduction]<br>
|
| 18 |
+
When using Huggingface's Serverless Inference API for a conversation, by default 100 new tokens are output and a cache is used.<br>
|
| 19 |
+
This API changes these two default settings, and other parameters are consistent with the official API.<br>
|
| 20 |
+
<br>
|
| 21 |
+
[How to use]<br>
|
| 22 |
+
1. <a target="_blank" href="https://huggingface.co/settings/tokens/new">Create a token</a> with the "Make calls to the serverless Inference API" permission as an API key.<br>
|
| 23 |
+
2. Set the Base URL of the OpenAI compatible client to "https://tastypear-sia-chat-adapter.hf.space/api".<br>
|
| 24 |
+
3. Use the full name of the model (e.g. mistralai/Mistral-Nemo-Instruct-2407)<br>
|
| 25 |
+
<br>
|
| 26 |
+
[Supported models]<br>
|
| 27 |
+
Most of the available models can be found <a target="_blank" href="https://huggingface.co/models?inference=warm&other=text-generation-inference">HERE</a>.<br>
|
| 28 |
+
Some "cold" models may also be supported (e.g. meta-llama/Meta-Llama-3.1-405B-Instruct), please test it yourself.<br>
|
| 29 |
+
Some models require a token created by a PRO user to use.<br>
|
| 30 |
+
<br>
|
| 31 |
+
[Avoid reaching the call limit]<br>
|
| 32 |
+
If you have multiple tokens, you can connect them with a semicolon (";") and the API will use a random one (e.g. "hf_aaaa;hf_bbbb;hf_...")<br>
|
| 33 |
+
</body>
|
| 34 |
+
</html>
|
| 35 |
+
'''
|
| 36 |
+
return render_template_string(template)
|
| 37 |
+
|
| 38 |
+
@app.route('/api/v1/chat/completions', methods=['POST'])
|
| 39 |
+
def proxy():
|
| 40 |
+
headers = dict(request.headers)
|
| 41 |
+
headers.pop('Host', None)
|
| 42 |
+
headers.pop('Content-Length', None)
|
| 43 |
+
keys = request.headers['Authorization'].split(' ')[1].split(';')
|
| 44 |
+
headers['Authorization'] = f'Bearer {random.choice(keys)}'
|
| 45 |
+
headers['X-Use-Cache'] = 'false'
|
| 46 |
+
|
| 47 |
+
json_data = request.get_json()
|
| 48 |
+
model = json_data['model']
|
| 49 |
+
chat_api = f"https://api-inference.huggingface.co/models/{model}/v1/chat/completions"
|
| 50 |
+
|
| 51 |
+
# Try to use the largest ctx
|
| 52 |
+
if not 'max_tokens' in json_data:
|
| 53 |
+
json_data['max_tokens'] = 2**32-1
|
| 54 |
+
json_data['json_mode'] = True
|
| 55 |
+
info = requests.post(chat_api, json=request.json, headers=headers, stream=False).text
|
| 56 |
+
max_ctx = int(info.split("<= ")[1].split(".")[0])
|
| 57 |
+
inputs = int(info.split("Given: ")[1].split("`")[0])
|
| 58 |
+
json_data['json_mode'] = False
|
| 59 |
+
json_data['max_tokens'] = max_ctx - inputs - 1
|
| 60 |
+
|
| 61 |
+
if not 'seed' in json_data:
|
| 62 |
+
json_data['seed'] = random.randint(1,2**32)
|
| 63 |
+
|
| 64 |
+
def generate():
|
| 65 |
+
with requests.post(chat_api, json=request.json, headers=headers, stream=True) as resp:
|
| 66 |
+
for chunk in resp.iter_content(chunk_size=1024):
|
| 67 |
+
if chunk:
|
| 68 |
+
yield chunk
|
| 69 |
+
|
| 70 |
+
return Response(stream_with_context(generate()), content_type='text/event-stream')
|
| 71 |
+
|
| 72 |
+
#import gevent.pywsgi
|
| 73 |
+
#from gevent import monkey;monkey.patch_all()
|
| 74 |
+
if __name__ == "__main__":
|
| 75 |
+
app.run(debug=True)
|
| 76 |
+
# gevent.pywsgi.WSGIServer((args.host, args.port), app).serve_forever()
|
requirements.txt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
requests
|
| 2 |
+
flask
|
| 3 |
+
gunicorn
|