Spaces:
Paused
Paused
Update models/local_llm.py
Browse files- models/local_llm.py +191 -6
models/local_llm.py
CHANGED
@@ -1,7 +1,192 @@
|
|
1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
LLM implementation using Hugging Face Inference Endpoint with OpenAI compatibility.
|
3 |
+
"""
|
4 |
+
import requests
|
5 |
+
import os
|
6 |
+
import json
|
7 |
+
import logging
|
8 |
+
from typing import Dict, List, Optional, Any
|
9 |
|
10 |
+
# Configure logging
|
11 |
+
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
|
12 |
+
logger = logging.getLogger(__name__)
|
13 |
+
|
14 |
+
# Endpoint configuration
|
15 |
+
HF_API_KEY = os.environ.get("HF_API_KEY", "")
|
16 |
+
ENDPOINT_URL = os.environ.get("ENDPOINT_URL", "https://cg01ow7izccjx1b2.us-east-1.aws.endpoints.huggingface.cloud/v1/chat/completions")
|
17 |
+
|
18 |
+
# Verify configuration
|
19 |
+
if not HF_API_KEY:
|
20 |
+
logger.warning("HF_API_KEY environment variable not set")
|
21 |
+
if not ENDPOINT_URL:
|
22 |
+
logger.warning("ENDPOINT_URL environment variable not set")
|
23 |
+
|
24 |
+
# Memory store for conversation history
|
25 |
+
conversation_memory: Dict[str, List[Dict[str, str]]] = {}
|
26 |
+
|
27 |
+
def run_llm(input_text: str, max_tokens: int = 512, temperature: float = 0.7) -> str:
|
28 |
+
"""
|
29 |
+
Process input text through HF Inference Endpoint.
|
30 |
+
|
31 |
+
Args:
|
32 |
+
input_text: User input to process
|
33 |
+
max_tokens: Maximum tokens to generate
|
34 |
+
temperature: Temperature for sampling (higher = more random)
|
35 |
+
|
36 |
+
Returns:
|
37 |
+
Generated response text
|
38 |
+
"""
|
39 |
+
headers = {
|
40 |
+
"Authorization": f"Bearer {HF_API_KEY}",
|
41 |
+
"Content-Type": "application/json"
|
42 |
+
}
|
43 |
+
|
44 |
+
# Format messages in OpenAI format
|
45 |
+
messages = [
|
46 |
+
{"role": "system", "content": "You are a helpful AI assistant for a telecom service. Answer questions clearly and concisely."},
|
47 |
+
{"role": "user", "content": input_text}
|
48 |
+
]
|
49 |
+
|
50 |
+
payload = {
|
51 |
+
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
|
52 |
+
"messages": messages,
|
53 |
+
"max_tokens": max_tokens,
|
54 |
+
"temperature": temperature
|
55 |
+
}
|
56 |
+
|
57 |
+
logger.debug(f"Sending request to endpoint with temperature={temperature}, max_tokens={max_tokens}")
|
58 |
+
|
59 |
+
try:
|
60 |
+
response = requests.post(ENDPOINT_URL, headers=headers, json=payload)
|
61 |
+
response.raise_for_status()
|
62 |
+
|
63 |
+
result = response.json()
|
64 |
+
response_text = result["choices"][0]["message"]["content"]
|
65 |
+
logger.debug(f"Generated response of {len(response_text)} characters")
|
66 |
+
return response_text
|
67 |
+
|
68 |
+
except requests.exceptions.RequestException as e:
|
69 |
+
error_msg = f"Error calling endpoint: {str(e)}"
|
70 |
+
if hasattr(e, 'response') and e.response is not None:
|
71 |
+
error_msg += f" - Status code: {e.response.status_code}, Response: {e.response.text}"
|
72 |
+
logger.error(error_msg)
|
73 |
+
return f"Error generating response: {str(e)}"
|
74 |
+
|
75 |
+
def run_llm_with_memory(input_text: str, session_id: str = "default", max_tokens: int = 512, temperature: float = 0.7) -> str:
|
76 |
+
"""
|
77 |
+
Process input with conversation memory.
|
78 |
+
|
79 |
+
Args:
|
80 |
+
input_text: User input to process
|
81 |
+
session_id: Unique identifier for conversation
|
82 |
+
max_tokens: Maximum tokens to generate
|
83 |
+
temperature: Temperature for sampling
|
84 |
+
|
85 |
+
Returns:
|
86 |
+
Generated response text
|
87 |
+
"""
|
88 |
+
# Initialize memory if needed
|
89 |
+
if session_id not in conversation_memory:
|
90 |
+
conversation_memory[session_id] = [
|
91 |
+
{"role": "system", "content": "You are a helpful AI assistant for a telecom service. Answer questions clearly and concisely."}
|
92 |
+
]
|
93 |
+
|
94 |
+
# Add current input to memory
|
95 |
+
conversation_memory[session_id].append({"role": "user", "content": input_text})
|
96 |
+
|
97 |
+
# Prepare the full conversation history
|
98 |
+
messages = conversation_memory[session_id].copy()
|
99 |
+
|
100 |
+
# Keep only the last 10 messages to avoid context length issues
|
101 |
+
if len(messages) > 10:
|
102 |
+
# Always keep the system message
|
103 |
+
messages = [messages[0]] + messages[-9:]
|
104 |
+
|
105 |
+
headers = {
|
106 |
+
"Authorization": f"Bearer {HF_API_KEY}",
|
107 |
+
"Content-Type": "application/json"
|
108 |
+
}
|
109 |
+
|
110 |
+
payload = {
|
111 |
+
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
|
112 |
+
"messages": messages,
|
113 |
+
"max_tokens": max_tokens,
|
114 |
+
"temperature": temperature
|
115 |
+
}
|
116 |
+
|
117 |
+
logger.debug(f"Sending memory-based request for session {session_id}")
|
118 |
+
|
119 |
+
try:
|
120 |
+
response = requests.post(ENDPOINT_URL, headers=headers, json=payload)
|
121 |
+
response.raise_for_status()
|
122 |
+
|
123 |
+
result = response.json()
|
124 |
+
response_text = result["choices"][0]["message"]["content"]
|
125 |
+
|
126 |
+
# Save response to memory
|
127 |
+
conversation_memory[session_id].append({"role": "assistant", "content": response_text})
|
128 |
+
|
129 |
+
return response_text
|
130 |
+
|
131 |
+
except requests.exceptions.RequestException as e:
|
132 |
+
error_msg = f"Error calling endpoint: {str(e)}"
|
133 |
+
if hasattr(e, 'response') and e.response is not None:
|
134 |
+
error_msg += f" - Status code: {e.response.status_code}, Response: {e.response.text}"
|
135 |
+
logger.error(error_msg)
|
136 |
+
return f"Error generating response: {str(e)}"
|
137 |
+
|
138 |
+
def clear_memory(session_id: str = "default") -> bool:
|
139 |
+
"""
|
140 |
+
Clear conversation memory for a specific session.
|
141 |
+
|
142 |
+
Args:
|
143 |
+
session_id: Unique identifier for conversation
|
144 |
+
"""
|
145 |
+
if session_id in conversation_memory:
|
146 |
+
conversation_memory[session_id] = [
|
147 |
+
{"role": "system", "content": "You are a helpful AI assistant for a telecom service. Answer questions clearly and concisely."}
|
148 |
+
]
|
149 |
+
return True
|
150 |
+
return False
|
151 |
+
|
152 |
+
def get_memory_sessions() -> List[str]:
|
153 |
+
"""
|
154 |
+
Get list of active memory sessions.
|
155 |
+
|
156 |
+
Returns:
|
157 |
+
List of session IDs
|
158 |
+
"""
|
159 |
+
return list(conversation_memory.keys())
|
160 |
+
|
161 |
+
def get_model_info() -> Dict[str, Any]:
|
162 |
+
"""
|
163 |
+
Get information about the connected model endpoint.
|
164 |
+
|
165 |
+
Returns:
|
166 |
+
Dictionary with endpoint information
|
167 |
+
"""
|
168 |
+
return {
|
169 |
+
"endpoint_url": ENDPOINT_URL,
|
170 |
+
"memory_sessions": len(conversation_memory),
|
171 |
+
"model_type": "Meta-Llama-3.1-8B-Instruct (Inference Endpoint)"
|
172 |
+
}
|
173 |
+
|
174 |
+
def test_endpoint() -> Dict[str, Any]:
|
175 |
+
"""
|
176 |
+
Test the endpoint connection.
|
177 |
+
|
178 |
+
Returns:
|
179 |
+
Status information
|
180 |
+
"""
|
181 |
+
try:
|
182 |
+
response = run_llm("Hello, this is a test message. Please respond with a short greeting.")
|
183 |
+
return {
|
184 |
+
"status": "connected",
|
185 |
+
"message": "Successfully connected to endpoint",
|
186 |
+
"sample_response": response[:50] + "..." if len(response) > 50 else response
|
187 |
+
}
|
188 |
+
except Exception as e:
|
189 |
+
return {
|
190 |
+
"status": "error",
|
191 |
+
"message": f"Failed to connect to endpoint: {str(e)}"
|
192 |
+
}
|