Spaces:
Paused
Paused
Update serve.py
Browse files
serve.py
CHANGED
|
@@ -1,117 +1,132 @@
|
|
| 1 |
-
# coding:utf-8
|
| 2 |
-
import argparse
|
| 3 |
-
import requests
|
| 4 |
-
import json
|
| 5 |
-
import os
|
| 6 |
-
from cerebras import CerebrasUnofficial
|
| 7 |
-
from flask import Flask, request, Response, stream_with_context, jsonify
|
| 8 |
-
import sys
|
| 9 |
-
|
| 10 |
-
# -- Start of Config --
|
| 11 |
-
|
| 12 |
-
# Replace with your cerebras.ai session token found in the `authjs.session-token` cookie.
|
| 13 |
-
# Or you can `set AUTHJS_SESSION_TOKEN=authjs.session-token`
|
| 14 |
-
# This token is valid for one month.
|
| 15 |
-
authjs_session_token = '12345678-abcd-abcd-abcd-12345678abcd'
|
| 16 |
-
|
| 17 |
-
# Replace with any string you wish like `my-api-key`.
|
| 18 |
-
# Or you can `set SERVER_API_KEY=my-api-key`
|
| 19 |
-
# You should set it to update the session token in the future.
|
| 20 |
-
server_api_key = 'my-api-key'
|
| 21 |
-
|
| 22 |
-
# -- End of Config --
|
| 23 |
-
|
| 24 |
-
sys.tracebacklimit = 0
|
| 25 |
-
|
| 26 |
-
authjs_session_token = os.environ.get('AUTHJS_SESSION_TOKEN', authjs_session_token)
|
| 27 |
-
server_api_key = os.environ.get('SERVER_API_KEY', server_api_key)
|
| 28 |
-
print(f'Using the cookie: authjs.session-token={authjs_session_token}')
|
| 29 |
-
print(f'Your api key: {server_api_key}')
|
| 30 |
-
|
| 31 |
-
cerebras_ai = CerebrasUnofficial(authjs_session_token)
|
| 32 |
-
|
| 33 |
-
app = Flask(__name__)
|
| 34 |
-
app.json.sort_keys = False
|
| 35 |
-
parser = argparse.ArgumentParser(description='Cerebras.AI API
|
| 36 |
-
parser.add_argument('--host', type=str, help='Set the ip address.(default: 0.0.0.0)', default='0.0.0.0')
|
| 37 |
-
parser.add_argument('--port', type=int, help='Set the port.(default: 7860)', default=7860)
|
| 38 |
-
args = parser.parse_args()
|
| 39 |
-
|
| 40 |
-
class Provider:
|
| 41 |
-
key = ''
|
| 42 |
-
max_tokens = None
|
| 43 |
-
api_url = ''
|
| 44 |
-
|
| 45 |
-
def __init__(self, request_key, model):
|
| 46 |
-
self.request_key = request_key
|
| 47 |
-
self.model = model
|
| 48 |
-
self.init_request_info()
|
| 49 |
-
|
| 50 |
-
def init_request_info(self):
|
| 51 |
-
if self.request_key == server_api_key:
|
| 52 |
-
self.api_url = cerebras_ai.api_url
|
| 53 |
-
self.key = cerebras_ai.get_api_key()
|
| 54 |
-
|
| 55 |
-
@app.route('/api
|
| 56 |
-
@app.route('/
|
| 57 |
-
def
|
| 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 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# coding:utf-8
|
| 2 |
+
import argparse
|
| 3 |
+
import requests
|
| 4 |
+
import json
|
| 5 |
+
import os
|
| 6 |
+
from cerebras import CerebrasUnofficial
|
| 7 |
+
from flask import Flask, request, Response, stream_with_context, jsonify
|
| 8 |
+
import sys
|
| 9 |
+
|
| 10 |
+
# -- Start of Config --
|
| 11 |
+
|
| 12 |
+
# Replace with your cerebras.ai session token found in the `authjs.session-token` cookie.
|
| 13 |
+
# Or you can `set AUTHJS_SESSION_TOKEN=authjs.session-token`
|
| 14 |
+
# This token is valid for one month.
|
| 15 |
+
authjs_session_token = '12345678-abcd-abcd-abcd-12345678abcd'
|
| 16 |
+
|
| 17 |
+
# Replace with any string you wish like `my-api-key`.
|
| 18 |
+
# Or you can `set SERVER_API_KEY=my-api-key`
|
| 19 |
+
# You should set it to update the session token in the future.
|
| 20 |
+
server_api_key = 'my-api-key'
|
| 21 |
+
|
| 22 |
+
# -- End of Config --
|
| 23 |
+
|
| 24 |
+
sys.tracebacklimit = 0
|
| 25 |
+
|
| 26 |
+
authjs_session_token = os.environ.get('AUTHJS_SESSION_TOKEN', authjs_session_token)
|
| 27 |
+
server_api_key = os.environ.get('SERVER_API_KEY', server_api_key)
|
| 28 |
+
print(f'Using the cookie: authjs.session-token={authjs_session_token}')
|
| 29 |
+
print(f'Your api key: {server_api_key}')
|
| 30 |
+
|
| 31 |
+
cerebras_ai = CerebrasUnofficial(authjs_session_token)
|
| 32 |
+
|
| 33 |
+
app = Flask(__name__)
|
| 34 |
+
app.json.sort_keys = False
|
| 35 |
+
parser = argparse.ArgumentParser(description='Cerebras.AI API')
|
| 36 |
+
parser.add_argument('--host', type=str, help='Set the ip address.(default: 0.0.0.0)', default='0.0.0.0')
|
| 37 |
+
parser.add_argument('--port', type=int, help='Set the port.(default: 7860)', default=7860)
|
| 38 |
+
args = parser.parse_args()
|
| 39 |
+
|
| 40 |
+
class Provider:
|
| 41 |
+
key = ''
|
| 42 |
+
max_tokens = None
|
| 43 |
+
api_url = ''
|
| 44 |
+
|
| 45 |
+
def __init__(self, request_key, model):
|
| 46 |
+
self.request_key = request_key
|
| 47 |
+
self.model = model
|
| 48 |
+
self.init_request_info()
|
| 49 |
+
|
| 50 |
+
def init_request_info(self):
|
| 51 |
+
if self.request_key == server_api_key:
|
| 52 |
+
self.api_url = cerebras_ai.api_url
|
| 53 |
+
self.key = cerebras_ai.get_api_key()
|
| 54 |
+
|
| 55 |
+
@app.route('/api', methods=['GET', 'POST'])
|
| 56 |
+
@app.route('/', methods=['GET', 'POST'])
|
| 57 |
+
def index():
|
| 58 |
+
return f'''
|
| 59 |
+
renew/change token by visiting:<br>
|
| 60 |
+
{request.host_url}renew?key={{your server api key}}&token={{your Cerebras authjs_session_token}}<br>
|
| 61 |
+
<br>
|
| 62 |
+
Your interface:<br>
|
| 63 |
+
{request.host_url}v1/chat/completions OR<br>
|
| 64 |
+
{request.host_url}api/v1/chat/completions<br>
|
| 65 |
+
<br>
|
| 66 |
+
For more infomation by visiting:<br>
|
| 67 |
+
https://github.com/tastypear/CerebrasUnofficial
|
| 68 |
+
'''
|
| 69 |
+
|
| 70 |
+
@app.route('/api/renew', methods=['GET', 'POST'])
|
| 71 |
+
@app.route('/renew', methods=['GET', 'POST'])
|
| 72 |
+
def renew_token():
|
| 73 |
+
if server_api_key == request.args.get('key', ''):
|
| 74 |
+
request_token = request.args.get('token', '')
|
| 75 |
+
global cerebras_ai
|
| 76 |
+
cerebras_ai = CerebrasUnofficial(request_token)
|
| 77 |
+
return f'new authjs.session_token: {request_token}'
|
| 78 |
+
else:
|
| 79 |
+
raise Exception('invalid api key')
|
| 80 |
+
|
| 81 |
+
@app.route('/api/v1/models', methods=['GET', 'POST'])
|
| 82 |
+
@app.route('/v1/models', methods=['GET', 'POST'])
|
| 83 |
+
def model_list():
|
| 84 |
+
model_list = {
|
| 85 |
+
'object': 'list',
|
| 86 |
+
'data': [{
|
| 87 |
+
'id': 'llama3.1-8b',
|
| 88 |
+
'object': 'model',
|
| 89 |
+
'created': 1721692800,
|
| 90 |
+
'owned_by': 'Meta'
|
| 91 |
+
}, {
|
| 92 |
+
'id': 'llama-3.3-70b',
|
| 93 |
+
'object': 'model',
|
| 94 |
+
'created': 1733443200,
|
| 95 |
+
'owned_by': 'Meta'
|
| 96 |
+
}, {
|
| 97 |
+
'id': 'deepseek-r1-distill-llama-70b',
|
| 98 |
+
'object': 'model',
|
| 99 |
+
'created': 1733443200,
|
| 100 |
+
'owned_by': 'deepseek'
|
| 101 |
+
}]
|
| 102 |
+
}
|
| 103 |
+
return jsonify(model_list)
|
| 104 |
+
|
| 105 |
+
|
| 106 |
+
@app.route('/api/v1/chat/completions', methods=['POST'])
|
| 107 |
+
@app.route('/v1/chat/completions', methods=['POST'])
|
| 108 |
+
def proxy():
|
| 109 |
+
request_key = request.headers['Authorization'].split(' ')[1]
|
| 110 |
+
if server_api_key != request_key:
|
| 111 |
+
raise Exception('invalid api key')
|
| 112 |
+
|
| 113 |
+
headers = dict(request.headers)
|
| 114 |
+
headers.pop('Host', None)
|
| 115 |
+
headers.pop('Content-Length', None)
|
| 116 |
+
|
| 117 |
+
headers['X-Use-Cache'] = 'false'
|
| 118 |
+
model = request.get_json()['model']
|
| 119 |
+
provider = Provider(request_key, model)
|
| 120 |
+
headers['Authorization'] = f'Bearer {provider.key}'
|
| 121 |
+
chat_api = f'{provider.api_url}/v1/chat/completions'
|
| 122 |
+
|
| 123 |
+
def generate():
|
| 124 |
+
with requests.post(chat_api, json=request.json, headers=headers, stream=True) as resp:
|
| 125 |
+
for chunk in resp.iter_content(chunk_size=1024):
|
| 126 |
+
if chunk:
|
| 127 |
+
chunk_str = chunk.decode('utf-8')
|
| 128 |
+
yield chunk_str
|
| 129 |
+
return Response(stream_with_context(generate()), content_type='text/event-stream')
|
| 130 |
+
|
| 131 |
+
if __name__ == '__main__':
|
| 132 |
+
app.run(host=args.host, port=args.port, debug=True)
|