Spaces:
Sleeping
Sleeping
tianlong12
commited on
Commit
•
c2f8097
1
Parent(s):
c61a34f
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
from flask import Flask, request, Response, jsonify
|
3 |
+
import requests
|
4 |
+
import json
|
5 |
+
import random
|
6 |
+
from helper import create_jwt, github_username_zed_userid_list
|
7 |
+
|
8 |
+
app = Flask(__name__)
|
9 |
+
|
10 |
+
@app.route('/hf/v1/chat/completions', methods=['POST'])
|
11 |
+
def chat():
|
12 |
+
# Get the payload from the request
|
13 |
+
payload = request.json
|
14 |
+
|
15 |
+
# Get the model from the payload, defaulting to "claude-3-5-sonnet-20240620"
|
16 |
+
model = payload.get('model', 'claude-3-5-sonnet-20240620')
|
17 |
+
|
18 |
+
# Prepare the request for the LLM API
|
19 |
+
url = "https://llm.zed.dev/completion?"
|
20 |
+
|
21 |
+
llm_payload = {
|
22 |
+
"provider": "anthropic",
|
23 |
+
"model": model,
|
24 |
+
"provider_request": {
|
25 |
+
"model": model,
|
26 |
+
"max_tokens": payload.get('max_tokens', 8192),
|
27 |
+
"temperature": payload.get('temperature', 0),
|
28 |
+
"top_p": payload.get('top_p', 0.7),
|
29 |
+
"messages": payload['messages'],
|
30 |
+
"stream": payload.get('stream', False),
|
31 |
+
"system": payload.get('system', "")
|
32 |
+
}
|
33 |
+
}
|
34 |
+
|
35 |
+
github_username, zed_user_id = random.choice(github_username_zed_userid_list)
|
36 |
+
jwt = create_jwt(github_username, zed_user_id)
|
37 |
+
|
38 |
+
headers = {
|
39 |
+
'Host': 'llm.zed.dev',
|
40 |
+
'accept': '*/*',
|
41 |
+
'content-type': 'application/json',
|
42 |
+
'authorization': f'Bearer {jwt}',
|
43 |
+
'user-agent': 'Zed/0.149.3 (macos; aarch64)'
|
44 |
+
}
|
45 |
+
|
46 |
+
# Get proxy from environment variable
|
47 |
+
proxy = os.environ.get('HTTP_PROXY', None)
|
48 |
+
proxies = {'http': proxy, 'https': proxy} if proxy else None
|
49 |
+
|
50 |
+
def generate():
|
51 |
+
with requests.post(url, headers=headers, json=llm_payload, stream=True, proxies=proxies) as response:
|
52 |
+
for chunk in response.iter_content(chunk_size=1024):
|
53 |
+
if chunk:
|
54 |
+
# Parse the chunk and format it as per OpenAI's streaming format
|
55 |
+
try:
|
56 |
+
data = json.loads(chunk.decode('utf-8'))
|
57 |
+
content = data.get('completion', '')
|
58 |
+
yield f"data: {json.dumps({'choices': [{'delta': {'content': content}}]})}\n\n"
|
59 |
+
except json.JSONDecodeError:
|
60 |
+
continue
|
61 |
+
yield "data: [DONE]\n\n"
|
62 |
+
|
63 |
+
if payload.get('stream', False):
|
64 |
+
return Response(generate(), content_type='text/event-stream')
|
65 |
+
else:
|
66 |
+
with requests.post(url, headers=headers, json=llm_payload, proxies=proxies) as response:
|
67 |
+
data = response.json()
|
68 |
+
return jsonify({
|
69 |
+
"id": "chatcmpl-" + os.urandom(12).hex(),
|
70 |
+
"object": "chat.completion",
|
71 |
+
"created": int(time.time()),
|
72 |
+
"model": model,
|
73 |
+
"choices": [{
|
74 |
+
"index": 0,
|
75 |
+
"message": {
|
76 |
+
"role": "assistant",
|
77 |
+
"content": data.get('completion', '')
|
78 |
+
},
|
79 |
+
"finish_reason": "stop"
|
80 |
+
}],
|
81 |
+
"usage": {
|
82 |
+
"prompt_tokens": -1, # We don't have this information
|
83 |
+
"completion_tokens": -1, # We don't have this information
|
84 |
+
"total_tokens": -1 # We don't have this information
|
85 |
+
}
|
86 |
+
})
|
87 |
+
|
88 |
+
if __name__ == '__main__':
|
89 |
+
app.run(debug=True)
|