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