Spaces:
Runtime error
Runtime error
Update main.py
Browse files
main.py
CHANGED
|
@@ -1,69 +1,69 @@
|
|
| 1 |
-
import random
|
| 2 |
-
import requests
|
| 3 |
-
from flask import Flask, request, jsonify, Response, stream_with_context, render_template_string
|
| 4 |
-
from mistral_common.protocol.instruct.messages import AssistantMessage, UserMessage, SystemMessage
|
| 5 |
-
from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
|
| 6 |
-
from mistral_common.protocol.instruct.request import ChatCompletionRequest
|
| 7 |
-
mt_v3 = MistralTokenizer.v3(is_tekken=True)
|
| 8 |
-
|
| 9 |
-
def calc_messages_tokens(json_data):
|
| 10 |
-
messages = json_data["messages"]
|
| 11 |
-
m_messages = []
|
| 12 |
-
for message in messages:
|
| 13 |
-
if message["role"] == "system":
|
| 14 |
-
m_messages.append(SystemMessage(content=message["content"]))
|
| 15 |
-
elif message["role"] == "user":
|
| 16 |
-
m_messages.append(UserMessage(content=message["content"]))
|
| 17 |
-
elif message["role"] == "assistant":
|
| 18 |
-
m_messages.append(AssistantMessage(content=message["content"]))
|
| 19 |
-
else:
|
| 20 |
-
continue
|
| 21 |
-
tokens = mt_v3.encode_chat_completion(ChatCompletionRequest(messages=m_messages)).tokens
|
| 22 |
-
return len(tokens) + len(m_messages)
|
| 23 |
-
|
| 24 |
-
app = Flask(__name__)
|
| 25 |
-
|
| 26 |
-
@app.route('/', methods=['GET'])
|
| 27 |
-
def index():
|
| 28 |
-
template = '''
|
| 29 |
-
<html>
|
| 30 |
-
<head>
|
| 31 |
-
<title>Mistral-Nemo Chat API</title>
|
| 32 |
-
</head>
|
| 33 |
-
<body>
|
| 34 |
-
<h1>Mistral-Nemo OpenAI Compatible API</h1>
|
| 35 |
-
<li>1. Create your key <a href="https://huggingface.co/settings/tokens/new">[here]</a>
|
| 36 |
-
<li>2. Set
|
| 37 |
-
If you have multiple keys, you can concatenate them with a semicolon (`;`) to use them randomly, e.g., `hf_aaaa;hf_bbbb;hf_...`
|
| 38 |
-
</body>
|
| 39 |
-
</html>
|
| 40 |
-
'''
|
| 41 |
-
return render_template_string(template)
|
| 42 |
-
|
| 43 |
-
@app.route('/api/v1/chat/completions', methods=['POST'])
|
| 44 |
-
def proxy():
|
| 45 |
-
headers = dict(request.headers)
|
| 46 |
-
headers.pop('Host', None)
|
| 47 |
-
headers.pop('Content-Length', None)
|
| 48 |
-
keys = request.headers['Authorization'].split(' ')[1].split(';')
|
| 49 |
-
headers['Authorization'] = f'Bearer {random.choice(keys)}'
|
| 50 |
-
|
| 51 |
-
json_data = request.get_json()
|
| 52 |
-
|
| 53 |
-
# Avoid using cache
|
| 54 |
-
json_data["messages"][-1]['content'] = ' '*random.randint(1, 20)+json_data["messages"][-1]['content']
|
| 55 |
-
|
| 56 |
-
# Use the largest ctx
|
| 57 |
-
json_data['max_tokens'] = 32768 - calc_messages_tokens(json_data)
|
| 58 |
-
|
| 59 |
-
json_data['json_mode'] = False
|
| 60 |
-
model = json_data['model']
|
| 61 |
-
|
| 62 |
-
def generate():
|
| 63 |
-
model = 'mistralai/Mistral-Nemo-Instruct-2407'
|
| 64 |
-
with requests.post(f"https://api-inference.huggingface.co/models/{model}/v1/chat/completions", json=request.json, headers=headers, stream=True) as resp:
|
| 65 |
-
for chunk in resp.iter_content(chunk_size=1024):
|
| 66 |
-
if chunk:
|
| 67 |
-
yield chunk
|
| 68 |
-
|
| 69 |
return Response(stream_with_context(generate()), content_type='text/event-stream')
|
|
|
|
| 1 |
+
import random
|
| 2 |
+
import requests
|
| 3 |
+
from flask import Flask, request, jsonify, Response, stream_with_context, render_template_string
|
| 4 |
+
from mistral_common.protocol.instruct.messages import AssistantMessage, UserMessage, SystemMessage
|
| 5 |
+
from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
|
| 6 |
+
from mistral_common.protocol.instruct.request import ChatCompletionRequest
|
| 7 |
+
mt_v3 = MistralTokenizer.v3(is_tekken=True)
|
| 8 |
+
|
| 9 |
+
def calc_messages_tokens(json_data):
|
| 10 |
+
messages = json_data["messages"]
|
| 11 |
+
m_messages = []
|
| 12 |
+
for message in messages:
|
| 13 |
+
if message["role"] == "system":
|
| 14 |
+
m_messages.append(SystemMessage(content=message["content"]))
|
| 15 |
+
elif message["role"] == "user":
|
| 16 |
+
m_messages.append(UserMessage(content=message["content"]))
|
| 17 |
+
elif message["role"] == "assistant":
|
| 18 |
+
m_messages.append(AssistantMessage(content=message["content"]))
|
| 19 |
+
else:
|
| 20 |
+
continue
|
| 21 |
+
tokens = mt_v3.encode_chat_completion(ChatCompletionRequest(messages=m_messages)).tokens
|
| 22 |
+
return len(tokens) + len(m_messages)
|
| 23 |
+
|
| 24 |
+
app = Flask(__name__)
|
| 25 |
+
|
| 26 |
+
@app.route('/', methods=['GET'])
|
| 27 |
+
def index():
|
| 28 |
+
template = '''
|
| 29 |
+
<html>
|
| 30 |
+
<head>
|
| 31 |
+
<title>Mistral-Nemo Chat API</title>
|
| 32 |
+
</head>
|
| 33 |
+
<body>
|
| 34 |
+
<h1>Mistral-Nemo OpenAI Compatible API</h1>
|
| 35 |
+
<li>1. Create your key <a href="https://huggingface.co/settings/tokens/new">[here]</a> with "serverless Inference API" permission selected.</li>
|
| 36 |
+
<li>2. Set "https://tastypear-mistral-nemo-chat.hf.space/api" as the domain in the client configuration.</li>
|
| 37 |
+
If you have multiple keys, you can concatenate them with a semicolon (`;`) to use them randomly, e.g., `hf_aaaa;hf_bbbb;hf_...`
|
| 38 |
+
</body>
|
| 39 |
+
</html>
|
| 40 |
+
'''
|
| 41 |
+
return render_template_string(template)
|
| 42 |
+
|
| 43 |
+
@app.route('/api/v1/chat/completions', methods=['POST'])
|
| 44 |
+
def proxy():
|
| 45 |
+
headers = dict(request.headers)
|
| 46 |
+
headers.pop('Host', None)
|
| 47 |
+
headers.pop('Content-Length', None)
|
| 48 |
+
keys = request.headers['Authorization'].split(' ')[1].split(';')
|
| 49 |
+
headers['Authorization'] = f'Bearer {random.choice(keys)}'
|
| 50 |
+
|
| 51 |
+
json_data = request.get_json()
|
| 52 |
+
|
| 53 |
+
# Avoid using cache
|
| 54 |
+
json_data["messages"][-1]['content'] = ' '*random.randint(1, 20)+json_data["messages"][-1]['content']
|
| 55 |
+
|
| 56 |
+
# Use the largest ctx
|
| 57 |
+
json_data['max_tokens'] = 32768 - calc_messages_tokens(json_data)
|
| 58 |
+
|
| 59 |
+
json_data['json_mode'] = False
|
| 60 |
+
model = json_data['model']
|
| 61 |
+
|
| 62 |
+
def generate():
|
| 63 |
+
model = 'mistralai/Mistral-Nemo-Instruct-2407'
|
| 64 |
+
with requests.post(f"https://api-inference.huggingface.co/models/{model}/v1/chat/completions", json=request.json, headers=headers, stream=True) as resp:
|
| 65 |
+
for chunk in resp.iter_content(chunk_size=1024):
|
| 66 |
+
if chunk:
|
| 67 |
+
yield chunk
|
| 68 |
+
|
| 69 |
return Response(stream_with_context(generate()), content_type='text/event-stream')
|