Spaces:
Sleeping
Sleeping
Upload 3 files
Browse files- core/llm.py +32 -0
- core/memory.py +15 -0
- core/utils.py +64 -0
core/llm.py
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
from dotenv import load_dotenv
|
| 3 |
+
from langchain_groq import ChatGroq
|
| 4 |
+
|
| 5 |
+
# Load environment variables
|
| 6 |
+
load_dotenv()
|
| 7 |
+
|
| 8 |
+
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
|
| 9 |
+
GROQ_MODEL = os.getenv("GROQ_MODEL", "llama2-70b-4096")
|
| 10 |
+
MAX_LENGTH = int(os.getenv("MAX_LENGTH", "512"))
|
| 11 |
+
TEMPERATURE = float(os.getenv("TEMPERATURE", "0.2"))
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
def create_llm():
|
| 15 |
+
if not GROQ_API_KEY or not GROQ_MODEL:
|
| 16 |
+
raise ValueError("GROQ_API_KEY and GROQ_MODEL must be set in the environment.")
|
| 17 |
+
return ChatGroq(
|
| 18 |
+
model=GROQ_MODEL,
|
| 19 |
+
api_key=GROQ_API_KEY,
|
| 20 |
+
temperature=TEMPERATURE,
|
| 21 |
+
max_tokens=MAX_LENGTH,
|
| 22 |
+
)
|
| 23 |
+
|
| 24 |
+
llm = None
|
| 25 |
+
try:
|
| 26 |
+
if GROQ_API_KEY and GROQ_MODEL:
|
| 27 |
+
llm = create_llm()
|
| 28 |
+
else:
|
| 29 |
+
print("LLM will not be initialized due to missing GROQ_API_KEY or GROQ_MODEL.")
|
| 30 |
+
except ValueError as e:
|
| 31 |
+
print(f"❌ Failed to initialize LLM: {e}")
|
| 32 |
+
llm = None
|
core/memory.py
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from langchain.memory import ConversationBufferWindowMemory
|
| 2 |
+
from typing import Dict
|
| 3 |
+
|
| 4 |
+
conversation_memories: Dict[str, ConversationBufferWindowMemory] = {}
|
| 5 |
+
|
| 6 |
+
def get_memory(session_id: str, max_history: int = 10) -> ConversationBufferWindowMemory:
|
| 7 |
+
"""Get or create conversation memory for a session."""
|
| 8 |
+
if session_id not in conversation_memories:
|
| 9 |
+
conversation_memories[session_id] = ConversationBufferWindowMemory(
|
| 10 |
+
k=max_history,
|
| 11 |
+
return_messages=True,
|
| 12 |
+
input_key="input",
|
| 13 |
+
output_key="output"
|
| 14 |
+
)
|
| 15 |
+
return conversation_memories[session_id]
|
core/utils.py
ADDED
|
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import re
|
| 2 |
+
from typing import Any, Dict, List
|
| 3 |
+
|
| 4 |
+
def classify_message(message: str) -> str:
|
| 5 |
+
"""Classify message type using LangChain approach."""
|
| 6 |
+
message_lower = message.lower().strip()
|
| 7 |
+
|
| 8 |
+
# Programming-related keywords
|
| 9 |
+
code_keywords = [
|
| 10 |
+
"write code", "python", "function", "class", "algorithm",
|
| 11 |
+
"programming", "code", "script", "loop", "variable", "import",
|
| 12 |
+
"def ", "return", "if ", "for ", "while ", "try:", "except:",
|
| 13 |
+
"list", "dict", "array", "dataframe", "numpy", "pandas"
|
| 14 |
+
]
|
| 15 |
+
|
| 16 |
+
# Conversational keywords
|
| 17 |
+
conversation_keywords = [
|
| 18 |
+
"hi", "hello", "how are you", "what's up", "good morning",
|
| 19 |
+
"good afternoon", "good evening", "bye", "goodbye", "thank you",
|
| 20 |
+
"thanks", "help", "who are you", "what can you do"
|
| 21 |
+
]
|
| 22 |
+
|
| 23 |
+
# Check for code-related content
|
| 24 |
+
if any(keyword in message_lower for keyword in code_keywords):
|
| 25 |
+
return "code"
|
| 26 |
+
|
| 27 |
+
# Check for conversational content
|
| 28 |
+
if any(keyword in message_lower for keyword in conversation_keywords):
|
| 29 |
+
return "conversation"
|
| 30 |
+
|
| 31 |
+
# Check for question marks or short messages
|
| 32 |
+
if message_lower.endswith("?") or len(message_lower.split()) <= 5:
|
| 33 |
+
return "conversation"
|
| 34 |
+
|
| 35 |
+
# Default to conversation for ambiguous cases
|
| 36 |
+
return "conversation"
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
def process_response(raw_response: str, response_type: str) -> Dict[str, Any]:
|
| 40 |
+
"""Process and format the model's response."""
|
| 41 |
+
if response_type == "conversation":
|
| 42 |
+
return {"response": raw_response.strip()}
|
| 43 |
+
elif response_type == "code":
|
| 44 |
+
code_match = re.search(r"```python\n(.*?)\n```", raw_response, re.DOTALL)
|
| 45 |
+
if code_match:
|
| 46 |
+
return {"generated_code": code_match.group(1).strip()}
|
| 47 |
+
return {"generated_code": raw_response.strip()}
|
| 48 |
+
elif response_type == "explanation":
|
| 49 |
+
explanation = re.sub(r"```python\n.*?\n```", "", raw_response, flags=re.DOTALL).strip()
|
| 50 |
+
return {"explanation": explanation}
|
| 51 |
+
else: # "both"
|
| 52 |
+
code = None
|
| 53 |
+
explanation = raw_response
|
| 54 |
+
code_match = re.search(r"```python\n(.*?)\n```", raw_response, re.DOTALL)
|
| 55 |
+
if code_match:
|
| 56 |
+
code = code_match.group(1).strip()
|
| 57 |
+
explanation = re.sub(r"```python\n.*?\n```", "", raw_response, flags=re.DOTALL).strip()
|
| 58 |
+
explanation = re.sub(r"\[EXPLANATION\]\s*", "", explanation, flags=re.IGNORECASE).strip()
|
| 59 |
+
explanation = re.sub(r"\[CODE\]\s*", "", explanation, flags=re.IGNORECASE).strip()
|
| 60 |
+
return {
|
| 61 |
+
"response": raw_response,
|
| 62 |
+
"generated_code": code,
|
| 63 |
+
"explanation": explanation
|
| 64 |
+
}
|