Spaces:
Running
Running
Update main.py
Browse files
main.py
CHANGED
@@ -1,3 +1,4 @@
|
|
|
|
1 |
from fastapi import FastAPI, HTTPException, Depends, Security, BackgroundTasks
|
2 |
from fastapi.security import APIKeyHeader
|
3 |
from fastapi.responses import StreamingResponse
|
@@ -13,11 +14,15 @@ import time
|
|
13 |
from datetime import datetime, timedelta
|
14 |
import asyncio
|
15 |
import requests
|
16 |
-
from prompts import CODING_ASSISTANT_PROMPT, NEWS_ASSISTANT_PROMPT, generate_news_prompt
|
17 |
from fastapi_cache import FastAPICache
|
18 |
from fastapi_cache.backends.inmemory import InMemoryBackend
|
19 |
from fastapi_cache.decorator import cache
|
20 |
|
|
|
|
|
|
|
|
|
21 |
app = FastAPI()
|
22 |
|
23 |
API_KEY_NAME = "X-API-Key"
|
@@ -69,6 +74,7 @@ class NewsQueryModel(BaseModel):
|
|
69 |
|
70 |
@lru_cache()
|
71 |
def get_api_keys():
|
|
|
72 |
return {
|
73 |
"OPENROUTER_API_KEY": f"sk-or-v1-{os.environ['OPENROUTER_API_KEY']}",
|
74 |
"BRAVE_API_KEY": os.environ['BRAVE_API_KEY']
|
@@ -85,12 +91,16 @@ last_activity: Dict[str, float] = {}
|
|
85 |
encoding = tiktoken.encoding_for_model("gpt-3.5-turbo")
|
86 |
|
87 |
def limit_tokens(input_string, token_limit=6000):
|
|
|
88 |
return encoding.decode(encoding.encode(input_string)[:token_limit])
|
89 |
|
90 |
def calculate_tokens(msgs):
|
91 |
-
|
|
|
|
|
92 |
|
93 |
def chat_with_llama_stream(messages, model="gpt-3.5-turbo", max_llm_history=4, max_output_tokens=2500):
|
|
|
94 |
while calculate_tokens(messages) > (8000 - max_output_tokens):
|
95 |
if len(messages) > max_llm_history:
|
96 |
messages = [messages[0]] + messages[-max_llm_history:]
|
@@ -98,9 +108,11 @@ def chat_with_llama_stream(messages, model="gpt-3.5-turbo", max_llm_history=4, m
|
|
98 |
max_llm_history -= 1
|
99 |
if max_llm_history < 2:
|
100 |
error_message = "Token limit exceeded. Please shorten your input or start a new conversation."
|
|
|
101 |
raise HTTPException(status_code=400, detail=error_message)
|
102 |
|
103 |
try:
|
|
|
104 |
response = or_client.chat.completions.create(
|
105 |
model=model,
|
106 |
messages=messages,
|
@@ -115,20 +127,25 @@ def chat_with_llama_stream(messages, model="gpt-3.5-turbo", max_llm_history=4, m
|
|
115 |
full_response += content
|
116 |
yield content
|
117 |
|
|
|
118 |
# After streaming, add the full response to the conversation history
|
119 |
messages.append({"role": "assistant", "content": full_response})
|
120 |
except Exception as e:
|
|
|
121 |
raise HTTPException(status_code=500, detail=f"Error in model response: {str(e)}")
|
122 |
|
123 |
async def verify_api_key(api_key: str = Security(api_key_header)):
|
124 |
if api_key != API_KEY:
|
|
|
125 |
raise HTTPException(status_code=403, detail="Could not validate credentials")
|
|
|
126 |
return api_key
|
127 |
|
128 |
# SQLite setup
|
129 |
DB_PATH = '/app/data/conversations.db'
|
130 |
|
131 |
def init_db():
|
|
|
132 |
os.makedirs(os.path.dirname(DB_PATH), exist_ok=True)
|
133 |
conn = sqlite3.connect(DB_PATH)
|
134 |
c = conn.cursor()
|
@@ -141,18 +158,22 @@ def init_db():
|
|
141 |
timestamp DATETIME DEFAULT CURRENT_TIMESTAMP)''')
|
142 |
conn.commit()
|
143 |
conn.close()
|
|
|
144 |
|
145 |
init_db()
|
146 |
|
147 |
def update_db(user_id, conversation_id, message, response):
|
|
|
148 |
conn = sqlite3.connect(DB_PATH)
|
149 |
c = conn.cursor()
|
150 |
c.execute('''INSERT INTO conversations (user_id, conversation_id, message, response)
|
151 |
VALUES (?, ?, ?, ?)''', (user_id, conversation_id, message, response))
|
152 |
conn.commit()
|
153 |
conn.close()
|
|
|
154 |
|
155 |
async def clear_inactive_conversations():
|
|
|
156 |
while True:
|
157 |
current_time = time.time()
|
158 |
inactive_convos = [conv_id for conv_id, last_time in last_activity.items()
|
@@ -162,27 +183,18 @@ async def clear_inactive_conversations():
|
|
162 |
del conversations[conv_id]
|
163 |
if conv_id in last_activity:
|
164 |
del last_activity[conv_id]
|
|
|
165 |
await asyncio.sleep(60) # Check every minute
|
166 |
|
167 |
@app.on_event("startup")
|
168 |
async def startup_event():
|
|
|
169 |
FastAPICache.init(InMemoryBackend(), prefix="fastapi-cache")
|
170 |
asyncio.create_task(clear_inactive_conversations())
|
171 |
|
172 |
@app.post("/coding-assistant")
|
173 |
async def coding_assistant(query: QueryModel, background_tasks: BackgroundTasks, api_key: str = Depends(verify_api_key)):
|
174 |
-
""
|
175 |
-
Coding assistant endpoint that provides programming help based on user queries.
|
176 |
-
Available models:
|
177 |
-
- meta-llama/llama-3-70b-instruct (default)
|
178 |
-
- anthropic/claude-3.5-sonnet
|
179 |
-
- deepseek/deepseek-coder
|
180 |
-
- anthropic/claude-3-haiku
|
181 |
-
- openai/gpt-3.5-turbo-instruct
|
182 |
-
- qwen/qwen-72b-chat
|
183 |
-
- google/gemma-2-27b-it
|
184 |
-
Requires API Key authentication via X-API-Key header.
|
185 |
-
"""
|
186 |
if query.conversation_id not in conversations:
|
187 |
conversations[query.conversation_id] = [
|
188 |
{"role": "system", "content": "You are a helpful assistant proficient in coding tasks. Help the user in understanding and writing code."}
|
@@ -199,18 +211,16 @@ async def coding_assistant(query: QueryModel, background_tasks: BackgroundTasks,
|
|
199 |
for content in chat_with_llama_stream(limited_conversation, model=query.model_id):
|
200 |
full_response += content
|
201 |
yield content
|
|
|
202 |
background_tasks.add_task(update_db, query.user_id, query.conversation_id, query.user_query, full_response)
|
203 |
|
204 |
return StreamingResponse(process_response(), media_type="text/event-stream")
|
205 |
|
206 |
# New functions for news assistant
|
207 |
|
208 |
-
def
|
209 |
-
|
210 |
-
|
211 |
-
else:
|
212 |
-
url = "https://api.search.brave.com/res/v1/news/search"
|
213 |
-
|
214 |
headers = {
|
215 |
"Accept": "application/json",
|
216 |
"Accept-Encoding": "gzip",
|
@@ -220,37 +230,33 @@ def internet_search(query, type = "web", num_results=20):
|
|
220 |
|
221 |
response = requests.get(url, headers=headers, params=params)
|
222 |
|
223 |
-
if response.status_code
|
224 |
-
|
225 |
-
|
226 |
-
|
227 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
228 |
else:
|
229 |
-
|
230 |
-
|
231 |
-
|
232 |
-
for item in search_data:
|
233 |
-
if not item.get("extra_snippets"):
|
234 |
-
continue
|
235 |
-
|
236 |
-
result = {
|
237 |
-
"title": item["title"],
|
238 |
-
"snippet": item["extra_snippets"][0],
|
239 |
-
"last_updated": item.get("age", "")
|
240 |
-
}
|
241 |
-
processed_results.append(result)
|
242 |
-
|
243 |
-
return processed_results[:num_results]
|
244 |
|
245 |
@lru_cache(maxsize=100)
|
246 |
-
def
|
247 |
-
|
248 |
-
|
249 |
|
250 |
def analyze_news(query):
|
251 |
-
|
|
|
252 |
|
253 |
if not news_data:
|
|
|
254 |
return "Failed to fetch news data.", []
|
255 |
|
256 |
# Prepare the prompt for the AI
|
@@ -262,72 +268,26 @@ def analyze_news(query):
|
|
262 |
{"role": "user", "content": prompt}
|
263 |
]
|
264 |
|
|
|
265 |
return messages
|
266 |
|
267 |
@app.post("/news-assistant")
|
268 |
async def news_assistant(query: NewsQueryModel, api_key: str = Depends(verify_api_key)):
|
269 |
-
""
|
270 |
-
News assistant endpoint that provides summaries and analysis of recent news based on user queries.
|
271 |
-
Requires API Key authentication via X-API-Key header.
|
272 |
-
"""
|
273 |
messages = analyze_news(query.query)
|
274 |
|
275 |
if not messages:
|
|
|
276 |
raise HTTPException(status_code=500, detail="Failed to fetch news data")
|
277 |
|
278 |
def process_response():
|
279 |
for content in chat_with_llama_stream(messages, model=query.model_id):
|
280 |
yield content
|
281 |
-
#meta-llama/llama-3-70b-instruct google/gemini-pro-1.5
|
282 |
-
return StreamingResponse(process_response(), media_type="text/event-stream")
|
283 |
-
|
284 |
-
class SearchQueryModel(BaseModel):
|
285 |
-
query: str = Field(..., description="Search query")
|
286 |
-
model_id: ModelID = Field(
|
287 |
-
default="meta-llama/llama-3-70b-instruct",
|
288 |
-
description="ID of the model to use for response generation"
|
289 |
-
)
|
290 |
-
class Config:
|
291 |
-
schema_extra = {
|
292 |
-
"example": {
|
293 |
-
"query": "What are the latest advancements in quantum computing?",
|
294 |
-
"model_id": "meta-llama/llama-3-70b-instruct"
|
295 |
-
}
|
296 |
-
}
|
297 |
-
|
298 |
-
def analyze_search_results(query):
|
299 |
-
search_data = internet_search(query, type="web")
|
300 |
-
|
301 |
-
if not search_data:
|
302 |
-
return "Failed to fetch search data.", []
|
303 |
-
|
304 |
-
# Prepare the prompt for the AI
|
305 |
-
prompt = generate_search_prompt(query, search_data)
|
306 |
-
|
307 |
-
messages = [
|
308 |
-
{"role": "system", "content": SEARCH_ASSISTANT_PROMPT},
|
309 |
-
{"role": "user", "content": prompt}
|
310 |
-
]
|
311 |
-
|
312 |
-
return messages
|
313 |
-
|
314 |
-
@app.post("/search-assistant")
|
315 |
-
async def search_assistant(query: SearchQueryModel, api_key: str = Depends(verify_api_key)):
|
316 |
-
"""
|
317 |
-
Search assistant endpoint that provides summaries and analysis of web search results based on user queries.
|
318 |
-
Requires API Key authentication via X-API-Key header.
|
319 |
-
"""
|
320 |
-
messages = analyze_search_results(query.query)
|
321 |
-
|
322 |
-
if not messages:
|
323 |
-
raise HTTPException(status_code=500, detail="Failed to fetch search data")
|
324 |
-
|
325 |
-
def process_response():
|
326 |
-
for content in chat_with_llama_stream(messages, model=query.model_id):
|
327 |
-
yield content
|
328 |
|
|
|
329 |
return StreamingResponse(process_response(), media_type="text/event-stream")
|
330 |
|
331 |
if __name__ == "__main__":
|
332 |
import uvicorn
|
|
|
333 |
uvicorn.run(app, host="0.0.0.0", port=7860)
|
|
|
1 |
+
import logging
|
2 |
from fastapi import FastAPI, HTTPException, Depends, Security, BackgroundTasks
|
3 |
from fastapi.security import APIKeyHeader
|
4 |
from fastapi.responses import StreamingResponse
|
|
|
14 |
from datetime import datetime, timedelta
|
15 |
import asyncio
|
16 |
import requests
|
17 |
+
from prompts import CODING_ASSISTANT_PROMPT, NEWS_ASSISTANT_PROMPT, generate_news_prompt
|
18 |
from fastapi_cache import FastAPICache
|
19 |
from fastapi_cache.backends.inmemory import InMemoryBackend
|
20 |
from fastapi_cache.decorator import cache
|
21 |
|
22 |
+
# Set up logging
|
23 |
+
logging.basicConfig(level=logging.DEBUG, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
|
24 |
+
logger = logging.getLogger(__name__)
|
25 |
+
|
26 |
app = FastAPI()
|
27 |
|
28 |
API_KEY_NAME = "X-API-Key"
|
|
|
74 |
|
75 |
@lru_cache()
|
76 |
def get_api_keys():
|
77 |
+
logger.debug("Fetching API keys")
|
78 |
return {
|
79 |
"OPENROUTER_API_KEY": f"sk-or-v1-{os.environ['OPENROUTER_API_KEY']}",
|
80 |
"BRAVE_API_KEY": os.environ['BRAVE_API_KEY']
|
|
|
91 |
encoding = tiktoken.encoding_for_model("gpt-3.5-turbo")
|
92 |
|
93 |
def limit_tokens(input_string, token_limit=6000):
|
94 |
+
logger.debug(f"Limiting tokens for input string, token limit: {token_limit}")
|
95 |
return encoding.decode(encoding.encode(input_string)[:token_limit])
|
96 |
|
97 |
def calculate_tokens(msgs):
|
98 |
+
token_count = sum(len(encoding.encode(str(m))) for m in msgs)
|
99 |
+
logger.debug(f"Calculated token count: {token_count}")
|
100 |
+
return token_count
|
101 |
|
102 |
def chat_with_llama_stream(messages, model="gpt-3.5-turbo", max_llm_history=4, max_output_tokens=2500):
|
103 |
+
logger.info(f"Starting chat with model: {model}")
|
104 |
while calculate_tokens(messages) > (8000 - max_output_tokens):
|
105 |
if len(messages) > max_llm_history:
|
106 |
messages = [messages[0]] + messages[-max_llm_history:]
|
|
|
108 |
max_llm_history -= 1
|
109 |
if max_llm_history < 2:
|
110 |
error_message = "Token limit exceeded. Please shorten your input or start a new conversation."
|
111 |
+
logger.error(error_message)
|
112 |
raise HTTPException(status_code=400, detail=error_message)
|
113 |
|
114 |
try:
|
115 |
+
logger.debug("Sending request to OpenAI")
|
116 |
response = or_client.chat.completions.create(
|
117 |
model=model,
|
118 |
messages=messages,
|
|
|
127 |
full_response += content
|
128 |
yield content
|
129 |
|
130 |
+
logger.debug("Finished streaming response")
|
131 |
# After streaming, add the full response to the conversation history
|
132 |
messages.append({"role": "assistant", "content": full_response})
|
133 |
except Exception as e:
|
134 |
+
logger.error(f"Error in model response: {str(e)}")
|
135 |
raise HTTPException(status_code=500, detail=f"Error in model response: {str(e)}")
|
136 |
|
137 |
async def verify_api_key(api_key: str = Security(api_key_header)):
|
138 |
if api_key != API_KEY:
|
139 |
+
logger.warning("Invalid API key attempt")
|
140 |
raise HTTPException(status_code=403, detail="Could not validate credentials")
|
141 |
+
logger.debug("API key verified successfully")
|
142 |
return api_key
|
143 |
|
144 |
# SQLite setup
|
145 |
DB_PATH = '/app/data/conversations.db'
|
146 |
|
147 |
def init_db():
|
148 |
+
logger.info("Initializing database")
|
149 |
os.makedirs(os.path.dirname(DB_PATH), exist_ok=True)
|
150 |
conn = sqlite3.connect(DB_PATH)
|
151 |
c = conn.cursor()
|
|
|
158 |
timestamp DATETIME DEFAULT CURRENT_TIMESTAMP)''')
|
159 |
conn.commit()
|
160 |
conn.close()
|
161 |
+
logger.debug("Database initialized")
|
162 |
|
163 |
init_db()
|
164 |
|
165 |
def update_db(user_id, conversation_id, message, response):
|
166 |
+
logger.debug(f"Updating database for conversation {conversation_id}")
|
167 |
conn = sqlite3.connect(DB_PATH)
|
168 |
c = conn.cursor()
|
169 |
c.execute('''INSERT INTO conversations (user_id, conversation_id, message, response)
|
170 |
VALUES (?, ?, ?, ?)''', (user_id, conversation_id, message, response))
|
171 |
conn.commit()
|
172 |
conn.close()
|
173 |
+
logger.debug("Database updated successfully")
|
174 |
|
175 |
async def clear_inactive_conversations():
|
176 |
+
logger.info("Starting inactive conversation cleanup task")
|
177 |
while True:
|
178 |
current_time = time.time()
|
179 |
inactive_convos = [conv_id for conv_id, last_time in last_activity.items()
|
|
|
183 |
del conversations[conv_id]
|
184 |
if conv_id in last_activity:
|
185 |
del last_activity[conv_id]
|
186 |
+
logger.debug(f"Cleared {len(inactive_convos)} inactive conversations")
|
187 |
await asyncio.sleep(60) # Check every minute
|
188 |
|
189 |
@app.on_event("startup")
|
190 |
async def startup_event():
|
191 |
+
logger.info("Starting up FastAPI application")
|
192 |
FastAPICache.init(InMemoryBackend(), prefix="fastapi-cache")
|
193 |
asyncio.create_task(clear_inactive_conversations())
|
194 |
|
195 |
@app.post("/coding-assistant")
|
196 |
async def coding_assistant(query: QueryModel, background_tasks: BackgroundTasks, api_key: str = Depends(verify_api_key)):
|
197 |
+
logger.info(f"Received coding assistant request for user {query.user_id}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
198 |
if query.conversation_id not in conversations:
|
199 |
conversations[query.conversation_id] = [
|
200 |
{"role": "system", "content": "You are a helpful assistant proficient in coding tasks. Help the user in understanding and writing code."}
|
|
|
211 |
for content in chat_with_llama_stream(limited_conversation, model=query.model_id):
|
212 |
full_response += content
|
213 |
yield content
|
214 |
+
logger.debug(f"Finished processing response for conversation {query.conversation_id}")
|
215 |
background_tasks.add_task(update_db, query.user_id, query.conversation_id, query.user_query, full_response)
|
216 |
|
217 |
return StreamingResponse(process_response(), media_type="text/event-stream")
|
218 |
|
219 |
# New functions for news assistant
|
220 |
|
221 |
+
def fetch_news(query, num_results=20):
|
222 |
+
logger.info(f"Fetching news for query: {query}")
|
223 |
+
url = "https://api.search.brave.com/res/v1/news/search"
|
|
|
|
|
|
|
224 |
headers = {
|
225 |
"Accept": "application/json",
|
226 |
"Accept-Encoding": "gzip",
|
|
|
230 |
|
231 |
response = requests.get(url, headers=headers, params=params)
|
232 |
|
233 |
+
if response.status_code == 200:
|
234 |
+
news_data = response.json()
|
235 |
+
logger.debug(f"Fetched {len(news_data['results'])} news items")
|
236 |
+
return [
|
237 |
+
{
|
238 |
+
"title": item["title"],
|
239 |
+
"snippet": item["extra_snippets"][0] if "extra_snippets" in item and item["extra_snippets"] else "",
|
240 |
+
"last_updated": item.get("age", ""),
|
241 |
+
}
|
242 |
+
for item in news_data['results']
|
243 |
+
if "extra_snippets" in item and item["extra_snippets"]
|
244 |
+
][:num_results]
|
245 |
else:
|
246 |
+
logger.error(f"Failed to fetch news. Status code: {response.status_code}")
|
247 |
+
return []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
248 |
|
249 |
@lru_cache(maxsize=100)
|
250 |
+
def cached_fetch_news(query: str):
|
251 |
+
logger.debug(f"Fetching cached news for query: {query}")
|
252 |
+
return fetch_news(query)
|
253 |
|
254 |
def analyze_news(query):
|
255 |
+
logger.info(f"Analyzing news for query: {query}")
|
256 |
+
news_data = cached_fetch_news(query)
|
257 |
|
258 |
if not news_data:
|
259 |
+
logger.warning("No news data fetched")
|
260 |
return "Failed to fetch news data.", []
|
261 |
|
262 |
# Prepare the prompt for the AI
|
|
|
268 |
{"role": "user", "content": prompt}
|
269 |
]
|
270 |
|
271 |
+
logger.debug("News analysis prompt prepared")
|
272 |
return messages
|
273 |
|
274 |
@app.post("/news-assistant")
|
275 |
async def news_assistant(query: NewsQueryModel, api_key: str = Depends(verify_api_key)):
|
276 |
+
logger.info(f"Received news assistant request for query: {query.query}")
|
|
|
|
|
|
|
277 |
messages = analyze_news(query.query)
|
278 |
|
279 |
if not messages:
|
280 |
+
logger.error("Failed to fetch news data")
|
281 |
raise HTTPException(status_code=500, detail="Failed to fetch news data")
|
282 |
|
283 |
def process_response():
|
284 |
for content in chat_with_llama_stream(messages, model=query.model_id):
|
285 |
yield content
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
286 |
|
287 |
+
logger.debug("Starting to stream news assistant response")
|
288 |
return StreamingResponse(process_response(), media_type="text/event-stream")
|
289 |
|
290 |
if __name__ == "__main__":
|
291 |
import uvicorn
|
292 |
+
logger.info("Starting uvicorn server")
|
293 |
uvicorn.run(app, host="0.0.0.0", port=7860)
|