Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,349 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import sys
|
3 |
+
import uvicorn
|
4 |
+
from fastapi import FastAPI, Query
|
5 |
+
from fastapi.responses import HTMLResponse
|
6 |
+
from starlette.middleware.cors import CORSMiddleware
|
7 |
+
from datasets import load_dataset, list_datasets
|
8 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
9 |
+
from loguru import logger
|
10 |
+
import concurrent.futures
|
11 |
+
import psutil
|
12 |
+
import asyncio
|
13 |
+
import torch
|
14 |
+
from tenacity import retry, stop_after_attempt, wait_fixed
|
15 |
+
from huggingface_hub import HfApi, RepositoryNotFoundError
|
16 |
+
from dotenv import load_dotenv
|
17 |
+
|
18 |
+
load_dotenv()
|
19 |
+
|
20 |
+
HUGGINGFACE_TOKEN = os.getenv("HUGGINGFACE_TOKEN")
|
21 |
+
if not HUGGINGFACE_TOKEN:
|
22 |
+
logger.error("Hugging Face token not found. Please set the HUGGINGFACE_TOKEN environment variable.")
|
23 |
+
sys.exit(1)
|
24 |
+
|
25 |
+
datasets_dict = {}
|
26 |
+
example_usage_list = []
|
27 |
+
|
28 |
+
CACHE_DIR = os.path.expanduser("~/.cache/huggingface")
|
29 |
+
os.makedirs(CACHE_DIR, exist_ok=True)
|
30 |
+
os.environ["HF_HOME"] = CACHE_DIR
|
31 |
+
os.environ["HF_TOKEN"] = HUGGINGFACE_TOKEN
|
32 |
+
|
33 |
+
def initialize_model():
|
34 |
+
try:
|
35 |
+
logger.info("Initializing the base model and tokenizer.")
|
36 |
+
base_model_repo = "Yhhxhfh/test"
|
37 |
+
model = AutoModelForCausalLM.from_pretrained(base_model_repo, cache_dir=CACHE_DIR)
|
38 |
+
tokenizer = AutoTokenizer.from_pretrained(base_model_repo, cache_dir=CACHE_DIR)
|
39 |
+
if tokenizer.pad_token is None:
|
40 |
+
tokenizer.pad_token = tokenizer.eos_token
|
41 |
+
pipeline_instance = pipeline(
|
42 |
+
"text-generation",
|
43 |
+
model=model,
|
44 |
+
tokenizer=tokenizer,
|
45 |
+
device=0 if torch.cuda.is_available() else -1
|
46 |
+
)
|
47 |
+
logger.info("Model and tokenizer initialized successfully.")
|
48 |
+
return pipeline_instance
|
49 |
+
except Exception as e:
|
50 |
+
logger.error(f"Error initializing model and tokenizer: {e}", exc_info=True)
|
51 |
+
sys.exit(1)
|
52 |
+
|
53 |
+
pipeline_instance = initialize_model()
|
54 |
+
|
55 |
+
@retry(stop=stop_after_attempt(3), wait=wait_fixed(5))
|
56 |
+
def download_dataset(dataset_name):
|
57 |
+
try:
|
58 |
+
logger.info(f"Starting download for dataset: {dataset_name}")
|
59 |
+
datasets_dict[dataset_name] = load_dataset(dataset_name, trust_remote_code=True, cache_dir=CACHE_DIR)
|
60 |
+
create_example_usage(dataset_name)
|
61 |
+
except Exception as e:
|
62 |
+
logger.error(f"Error loading dataset {dataset_name}: {e}", exc_info=True)
|
63 |
+
raise
|
64 |
+
|
65 |
+
def upload_model_to_hub():
|
66 |
+
try:
|
67 |
+
api = HfApi()
|
68 |
+
model_repo = "Yhhxhfh/test"
|
69 |
+
try:
|
70 |
+
api.repo_info(repo_id=model_repo)
|
71 |
+
logger.info(f"Model repository {model_repo} already exists.")
|
72 |
+
except RepositoryNotFoundError:
|
73 |
+
api.create_repo(repo_id=model_repo, private=False, token=HUGGINGFACE_TOKEN)
|
74 |
+
logger.info(f"Created model repository {model_repo}.")
|
75 |
+
logger.info(f"Pushing the model and tokenizer to {model_repo}.")
|
76 |
+
pipeline_instance.model.push_to_hub(model_repo, use_auth_token=HUGGINGFACE_TOKEN)
|
77 |
+
pipeline_instance.tokenizer.push_to_hub(model_repo, use_auth_token=HUGGINGFACE_TOKEN)
|
78 |
+
logger.info(f"Successfully pushed the model and tokenizer to {model_repo}.")
|
79 |
+
except Exception as e:
|
80 |
+
logger.error(f"Error uploading model to Hugging Face Hub: {e}", exc_info=True)
|
81 |
+
|
82 |
+
def create_example_usage(dataset_name):
|
83 |
+
try:
|
84 |
+
logger.info(f"Creating example usage for dataset {dataset_name}")
|
85 |
+
example_prompts = [
|
86 |
+
"Translate the following catering menu from English to French:",
|
87 |
+
"Generate a catering menu for a wedding with vegetarian options:",
|
88 |
+
"Convert the following catering menu to a gluten-free version:",
|
89 |
+
"Provide a detailed catering menu for a corporate event including desserts:",
|
90 |
+
"Generate a children's birthday party catering menu with allergen-free items:"
|
91 |
+
]
|
92 |
+
examples = []
|
93 |
+
for prompt in example_prompts:
|
94 |
+
generated_text = pipeline_instance(prompt, max_length=50, num_return_sequences=1)[0]['generated_text']
|
95 |
+
examples.append({"prompt": prompt, "response": generated_text})
|
96 |
+
example_usage_list.append({"dataset_name": dataset_name, "examples": examples})
|
97 |
+
logger.info(f"Example usage created for dataset {dataset_name}")
|
98 |
+
except Exception as e:
|
99 |
+
logger.error(f"Error creating example usage for dataset {dataset_name}: {e}", exc_info=True)
|
100 |
+
|
101 |
+
def unify_datasets():
|
102 |
+
try:
|
103 |
+
logger.info("Starting to unify datasets")
|
104 |
+
unified_dataset = None
|
105 |
+
for dataset in datasets_dict.values():
|
106 |
+
if unified_dataset is None:
|
107 |
+
unified_dataset = dataset
|
108 |
+
else:
|
109 |
+
unified_dataset = unified_dataset.concatenate(dataset)
|
110 |
+
datasets_dict['unified'] = unified_dataset
|
111 |
+
logger.info("Datasets successfully unified.")
|
112 |
+
except Exception as e:
|
113 |
+
logger.error(f"Error unifying datasets: {e}", exc_info=True)
|
114 |
+
|
115 |
+
cpu_count = psutil.cpu_count(logical=False) or 1
|
116 |
+
memory_available_mb = psutil.virtual_memory().available / (1024 * 1024)
|
117 |
+
memory_per_download_mb = 100
|
118 |
+
memory_available = int(memory_available_mb / memory_per_download_mb)
|
119 |
+
gpu_count = torch.cuda.device_count()
|
120 |
+
max_concurrent_downloads = min(cpu_count, memory_available, gpu_count * 2 if gpu_count else cpu_count)
|
121 |
+
max_concurrent_downloads = max(1, max_concurrent_downloads)
|
122 |
+
max_concurrent_downloads = min(10, max_concurrent_downloads)
|
123 |
+
|
124 |
+
logger.info(f"Using up to {max_concurrent_downloads} concurrent workers for downloading datasets.")
|
125 |
+
|
126 |
+
executor = concurrent.futures.ThreadPoolExecutor(max_workers=max_concurrent_downloads)
|
127 |
+
|
128 |
+
async def download_and_process_datasets():
|
129 |
+
dataset_names = list_datasets()
|
130 |
+
logger.info(f"Found {len(dataset_names)} datasets to download.")
|
131 |
+
loop = asyncio.get_event_loop()
|
132 |
+
tasks = []
|
133 |
+
for dataset_name in dataset_names:
|
134 |
+
task = loop.run_in_executor(executor, download_dataset, dataset_name)
|
135 |
+
tasks.append(task)
|
136 |
+
await asyncio.gather(*tasks)
|
137 |
+
unify_datasets()
|
138 |
+
upload_model_to_hub()
|
139 |
+
|
140 |
+
async def main():
|
141 |
+
await download_and_process_datasets()
|
142 |
+
|
143 |
+
asyncio.run(main())
|
144 |
+
|
145 |
+
app = FastAPI()
|
146 |
+
|
147 |
+
app.add_middleware(
|
148 |
+
CORSMiddleware,
|
149 |
+
allow_origins=["*"],
|
150 |
+
allow_credentials=True,
|
151 |
+
allow_methods=["*"],
|
152 |
+
allow_headers=["*"]
|
153 |
+
)
|
154 |
+
|
155 |
+
message_history = []
|
156 |
+
|
157 |
+
@app.get('/')
|
158 |
+
async def index():
|
159 |
+
html_code = """
|
160 |
+
<!DOCTYPE html>
|
161 |
+
<html lang="en">
|
162 |
+
<head>
|
163 |
+
<meta charset="UTF-8">
|
164 |
+
<meta name="viewport" content="width=device-width, initial-scale=1.0">
|
165 |
+
<title>ChatGPT Chatbot</title>
|
166 |
+
<style>
|
167 |
+
body {
|
168 |
+
font-family: Arial, sans-serif;
|
169 |
+
margin: 0;
|
170 |
+
padding: 0;
|
171 |
+
background-color: #f4f4f4;
|
172 |
+
}
|
173 |
+
.container {
|
174 |
+
max-width: 800px;
|
175 |
+
margin: auto;
|
176 |
+
padding: 20px;
|
177 |
+
}
|
178 |
+
.chat-container {
|
179 |
+
background-color: #fff;
|
180 |
+
border-radius: 8px;
|
181 |
+
box-shadow: 0 0 10px rgba(0, 0, 0, 0.1);
|
182 |
+
overflow: hidden;
|
183 |
+
margin-bottom: 20px;
|
184 |
+
animation: fadeInUp 0.5s ease forwards;
|
185 |
+
display: flex;
|
186 |
+
flex-direction: column;
|
187 |
+
}
|
188 |
+
.chat-box {
|
189 |
+
flex: 1;
|
190 |
+
overflow-y: auto;
|
191 |
+
padding: 10px;
|
192 |
+
}
|
193 |
+
.chat-input {
|
194 |
+
width: calc(100% - 20px);
|
195 |
+
border: none;
|
196 |
+
border-top: 1px solid #ddd;
|
197 |
+
padding: 10px;
|
198 |
+
font-size: 16px;
|
199 |
+
outline: none;
|
200 |
+
}
|
201 |
+
.chat-input:focus {
|
202 |
+
border-top: 1px solid #007bff;
|
203 |
+
}
|
204 |
+
.user-message {
|
205 |
+
margin-bottom: 10px;
|
206 |
+
padding: 8px 12px;
|
207 |
+
border-radius: 8px;
|
208 |
+
background-color: #007bff;
|
209 |
+
color: #fff;
|
210 |
+
max-width: 70%;
|
211 |
+
word-wrap: break-word;
|
212 |
+
align-self: flex-end;
|
213 |
+
}
|
214 |
+
.bot-message {
|
215 |
+
margin-bottom: 10px;
|
216 |
+
padding: 8px 12px;
|
217 |
+
border-radius: 8px;
|
218 |
+
background-color: #4CAF50;
|
219 |
+
color: #fff;
|
220 |
+
max-width: 70%;
|
221 |
+
word-wrap: break-word;
|
222 |
+
}
|
223 |
+
.toggle-history {
|
224 |
+
text-align: center;
|
225 |
+
cursor: pointer;
|
226 |
+
color: #007bff;
|
227 |
+
margin-bottom: 10px;
|
228 |
+
}
|
229 |
+
.history-container {
|
230 |
+
display: none;
|
231 |
+
}
|
232 |
+
.history-container.show {
|
233 |
+
display: block;
|
234 |
+
}
|
235 |
+
.history-container .history-content {
|
236 |
+
max-height: 200px;
|
237 |
+
overflow-y: auto;
|
238 |
+
}
|
239 |
+
@keyframes fadeInUp {
|
240 |
+
from {
|
241 |
+
opacity: 0;
|
242 |
+
transform: translateY(20px);
|
243 |
+
}
|
244 |
+
to {
|
245 |
+
opacity: 1;
|
246 |
+
transform: translateY(0);
|
247 |
+
}
|
248 |
+
}
|
249 |
+
</style>
|
250 |
+
</head>
|
251 |
+
<body>
|
252 |
+
<div class="container">
|
253 |
+
<h1 style="text-align: center;">ChatGPT Chatbot</h1>
|
254 |
+
<div class="chat-container" id="chat-container">
|
255 |
+
<div class="chat-box" id="chat-box">
|
256 |
+
</div>
|
257 |
+
<input type="text" class="chat-input" id="user-input" placeholder="Type your message...">
|
258 |
+
<button onclick="retryLastMessage()">Retry Last Message</button>
|
259 |
+
</div>
|
260 |
+
<div class="toggle-history" onclick="toggleHistory()">Toggle History</div>
|
261 |
+
<div class="history-container" id="history-container">
|
262 |
+
<h2>Chat History</h2>
|
263 |
+
<div class="history-content" id="history-content"></div>
|
264 |
+
</div>
|
265 |
+
</div>
|
266 |
+
<script>
|
267 |
+
function toggleHistory() {
|
268 |
+
const historyContainer = document.getElementById('history-container');
|
269 |
+
historyContainer.classList.toggle('show');
|
270 |
+
}
|
271 |
+
|
272 |
+
function saveMessage(sender, message) {
|
273 |
+
const historyContent = document.getElementById('history-content');
|
274 |
+
const messageElement = document.createElement('div');
|
275 |
+
messageElement.className = `${sender}-message`;
|
276 |
+
messageElement.innerText = message;
|
277 |
+
historyContent.appendChild(messageElement);
|
278 |
+
}
|
279 |
+
|
280 |
+
function appendMessage(sender, message) {
|
281 |
+
const chatBox = document.getElementById('chat-box');
|
282 |
+
const messageElement = document.createElement('div');
|
283 |
+
messageElement.className = `${sender}-message`;
|
284 |
+
messageElement.innerText = message;
|
285 |
+
chatBox.appendChild(messageElement);
|
286 |
+
chatBox.scrollTop = chatBox.scrollHeight;
|
287 |
+
}
|
288 |
+
|
289 |
+
const chatContainer = document.getElementById('chat-container');
|
290 |
+
const chatBox = document.getElementById('chat-box');
|
291 |
+
const userInput = document.getElementById('user-input');
|
292 |
+
|
293 |
+
userInput.addEventListener('keyup', function(event) {
|
294 |
+
if (event.keyCode === 13) {
|
295 |
+
event.preventDefault();
|
296 |
+
sendMessage();
|
297 |
+
}
|
298 |
+
});
|
299 |
+
|
300 |
+
function sendMessage() {
|
301 |
+
const userMessage = userInput.value.trim();
|
302 |
+
if (userMessage === '') return;
|
303 |
+
|
304 |
+
saveMessage('user', userMessage);
|
305 |
+
appendMessage('user', userMessage);
|
306 |
+
userInput.value = '';
|
307 |
+
|
308 |
+
fetch(`/autocomplete?q=${encodeURIComponent(userMessage)}`)
|
309 |
+
.then(response => response.json())
|
310 |
+
.then(data => {
|
311 |
+
const botMessages = data.result;
|
312 |
+
botMessages.forEach(message => {
|
313 |
+
saveMessage('bot', message);
|
314 |
+
appendMessage('bot', message);
|
315 |
+
});
|
316 |
+
})
|
317 |
+
.catch(error => {
|
318 |
+
console.error('Error:', error);
|
319 |
+
});
|
320 |
+
}
|
321 |
+
|
322 |
+
function retryLastMessage() {
|
323 |
+
const lastUserMessage = document.querySelector('.user-message:last-of-type');
|
324 |
+
if (lastUserMessage) {
|
325 |
+
userInput.value = lastUserMessage.innerText;
|
326 |
+
sendMessage();
|
327 |
+
}
|
328 |
+
}
|
329 |
+
</script>
|
330 |
+
</body>
|
331 |
+
</html>
|
332 |
+
"""
|
333 |
+
return HTMLResponse(content=html_code, status_code=200)
|
334 |
+
|
335 |
+
@app.get('/autocomplete')
|
336 |
+
async def autocomplete(q: str = Query(..., title='query')):
|
337 |
+
global message_history
|
338 |
+
message_history.append(('user', q))
|
339 |
+
try:
|
340 |
+
response = pipeline_instance(q, max_length=50, num_return_sequences=1)[0]['generated_text']
|
341 |
+
logger.debug(f"Successfully autocomplete, q:{q}, res:{response}")
|
342 |
+
return {"result": [response]}
|
343 |
+
except Exception as e:
|
344 |
+
logger.error(f"Ignored error in autocomplete: {e}", exc_info=True)
|
345 |
+
return {"result": []}
|
346 |
+
|
347 |
+
if __name__ == '__main__':
|
348 |
+
port = int(os.getenv("PORT", 443))
|
349 |
+
uvicorn.run(app=app, host='0.0.0.0', port=port)
|