final

app.py

@@ -3,6 +3,7 @@ import logging
 from typing import Optional, List
 import asyncio
 from contextlib import asynccontextmanager
+import json
 
 import pymongo
 from fastapi import FastAPI, HTTPException
@@ -17,9 +18,7 @@ from langchain_core.messages import (
 )
 from langchain_core.prompts import PromptTemplate
 from langchain_groq import ChatGroq
-from langchain_core.output_parsers import PydanticOutputParser, StrOutputParser
-from langchain_core.chat_history import BaseChatMessageHistory
-from langchain_core.runnables.history import RunnableWithMessageHistory
+from langchain_core.output_parsers import StrOutputParser
 
 # -------------------------------------------------------------------
 # CONFIGURATION
@@ -46,7 +45,7 @@ history_collection = None
 # -------------------------------------------------------------------
 # CHAT HISTORY HANDLER (MongoDB)
 # -------------------------------------------------------------------
-class MongoDBChatMessageHistory(BaseChatMessageHistory):
+class MongoDBChatMessageHistory:
     """Chat message history stored in MongoDB."""
 
     def __init__(self, collection, session_id: str):
@@ -55,19 +54,18 @@ class MongoDBChatMessageHistory(BaseChatMessageHistory):
         self.collection.create_index("session_id")
 
     @property
-    def messages(self) -> List[BaseMessage]:
+    def messages(self) -> List[dict]:
         """Retrieve messages from MongoDB"""
         document = self.collection.find_one({"session_id": self.session_id})
         if document and "history" in document:
-            return messages_from_dict(document["history"])
+            return document["history"]
         return []
 
-    def add_message(self, message: BaseMessage) -> None:
+    def add_message(self, message: dict) -> None:
         """Append the message to MongoDB history"""
-        message_dict = message_to_dict(message)
         self.collection.update_one(
             {"session_id": self.session_id},
-            {"$push": {"history": message_dict}},
+            {"$push": {"history": message}},
             upsert=True,
         )
 
@@ -86,9 +84,9 @@ async def lifespan(app: FastAPI):
         # Initialize LLM
         llm = ChatGroq(
             groq_api_key=GROQ_API_KEY,
-            model_name="openai/gpt-oss-20b",
-            temperature=0,
-            max_tokens=600
+            model_name="mixtral-8x7b-32768",
+            temperature=0.3,
+            max_tokens=800
         )
         logger.info("ChatGroq LLM initialized successfully")
 
@@ -112,8 +110,8 @@ async def lifespan(app: FastAPI):
 # -------------------------------------------------------------------
 app = FastAPI(
     title="India Legal Consultation Service",
-    description="A bilingual (Hindi-English) legal consultation API powered by Groq and MongoDB.",
-    version="2.0.0",
+    description="A bilingual legal consultation API powered by Groq and MongoDB",
+    version="3.0.0",
     lifespan=lifespan
 )
 
@@ -129,19 +127,13 @@ app.add_middleware(
 # -------------------------------------------------------------------
 # PYDANTIC MODELS
 # -------------------------------------------------------------------
-class ConsultationResponse(BaseModel):
-    consultation: str
-    key_terms: Optional[str] = "general"
-
-consultation_parser = PydanticOutputParser(pydantic_object=ConsultationResponse)
-
 class QueryRequest(BaseModel):
     query: str
     user_id: str
 
-class ConsultationResponseWithHistory(BaseModel):
+class ConsultationResponse(BaseModel):
     response: str
-    history: List[str]
+    history: List[dict] = Field(default_factory=list)
 
 class HistoryResponse(BaseModel):
     history: List[dict] = Field(default_factory=list)
@@ -152,60 +144,67 @@ class DeleteResponse(BaseModel):
 
 
 # -------------------------------------------------------------------
-# LEGAL CONSULTATION PROMPT (FIXED - NO NEWLINE CHARS IN OUTPUT)
+# LEGAL CONSULTATION PROMPT (FIXED - NO JSON PARSING ISSUES)
 # -------------------------------------------------------------------
 consultation_prompt = PromptTemplate(
     input_variables=["query", "previous_messages"],
-    template="""You are an expert Indian legal advisor (behave like a senior advocate or legal consultant with deep understanding of Indian laws). Context: Current user query: "{query}" Previous conversation: "{previous_messages}" Instructions: 1. Analyze the user's question under Indian law context (IPC, CrPC, Indian Evidence Act, Constitution, Family Law, Contract Law, IT Act, etc.). 2. Provide a clear, practical, and lawful explanation that can help a normal citizen understand their rights and possible actions. 3. Include relevant sections or provisions from Indian law where applicable. 4. Explain in both Hindi and English (use short, readable paragraphs). 5. Keep the tone professional, supportive, and educational, not judgmental. 6. Avoid giving direct legal verdicts; instead, provide guidance and awareness. 7. Extract one main legal keyword or area (e.g., murder, divorce, contract breach, cybercrime, tenant rights, property dispute). Output: Strictly follow this JSON structure without newline characters: {format_instructions}""",
-    partial_variables={"format_instructions": consultation_parser.get_format_instructions()},
-)
+    template="""You are an expert Indian legal advisor with deep knowledge of Indian law. Analyze the following query and provide guidance based on Indian laws (IPC, CrPC, Indian Evidence Act, Constitution, Family Law, Contract Law, IT Act, etc.).
 
+User Query: {query}
 
-# -------------------------------------------------------------------
-# FUNCTIONS TO BUILD CHAINS
-# -------------------------------------------------------------------
-def get_consultation_chain():
-    if llm is None:
-        raise HTTPException(status_code=503, detail="LLM not initialized yet")
-    return consultation_prompt | llm | StrOutputParser()
-
+Previous Conversation: {previous_messages}
 
-def get_consultation_runnable():
-    consultation_chain = get_consultation_chain()
-    return RunnableWithMessageHistory(
-        consultation_chain,
-        lambda user_id: MongoDBChatMessageHistory(collection=history_collection, session_id=user_id),
-        input_messages_key="query",
-        history_messages_key="previous_messages",
-    )
+Instructions:
+1. Provide a clear, practical explanation of the legal issue under Indian law
+2. Include relevant law sections or provisions where applicable
+3. Explain in both Hindi and English using short readable paragraphs
+4. Keep tone professional, supportive, and educational
+5. Provide guidance and awareness, not direct legal verdicts
+6. Identify the main legal area involved (e.g., murder, divorce, contract, cybercrime, tenant rights, property dispute)
 
+Response format: Start with Hindi explanation, then English explanation. End with the legal area identified."""
+)
 
 # -------------------------------------------------------------------
 # API ROUTES
 # -------------------------------------------------------------------
-@app.post("/legal-consultation", response_model=ConsultationResponseWithHistory)
+@app.post("/legal-consultation", response_model=ConsultationResponse)
 def legal_consultation(request: QueryRequest):
     if llm is None or history_collection is None:
         raise HTTPException(status_code=503, detail="Service not ready. Try again later.")
 
     try:
-        consultation_runnable = get_consultation_runnable()
-
-        result = consultation_runnable.invoke(
-            {"query": request.query},
-            {"configurable": {"session_id": request.user_id}}
-        )
-
         history_obj = MongoDBChatMessageHistory(history_collection, request.user_id)
-        all_messages = history_obj.messages
-
-        return ConsultationResponseWithHistory(
+        messages = history_obj.messages
+        
+        # Format previous messages
+        previous_context = ""
+        if messages:
+            previous_context = " ".join([f"{msg.get('role', 'user')}: {msg.get('content', '')}" for msg in messages[-5:]])
+        
+        consultation_chain = consultation_prompt | llm | StrOutputParser()
+        
+        result = consultation_chain.invoke({
+            "query": request.query,
+            "previous_messages": previous_context
+        })
+
+        user_msg = {"role": "user", "content": request.query}
+        ai_msg = {"role": "assistant", "content": result}
+        
+        history_obj.add_message(user_msg)
+        history_obj.add_message(ai_msg)
+
+        # Return updated history
+        updated_messages = history_obj.messages
+        
+        return ConsultationResponse(
             response=result,
-            history=[msg.content for msg in all_messages]
+            history=updated_messages
         )
     except Exception as e:
         logger.error(f"Consultation failed: {e}")
-        raise HTTPException(status_code=500, detail=f"Consultation failed: {e}")
+        raise HTTPException(status_code=500, detail=f"Consultation failed: {str(e)}")
 
 
 @app.get("/get-session-history/{session_id}", response_model=HistoryResponse)
@@ -215,11 +214,10 @@ async def get_session_history(session_id: str):
     try:
         history_obj = MongoDBChatMessageHistory(history_collection, session_id)
         messages = history_obj.messages
-        history_dicts = [message_to_dict(msg) for msg in messages]
-        return HistoryResponse(history=history_dicts)
+        return HistoryResponse(history=messages)
     except Exception as e:
         logger.error(f"Failed to get history for {session_id}: {e}")
-        raise HTTPException(status_code=500, detail=f"Failed to retrieve history: {e}")
+        raise HTTPException(status_code=500, detail=f"Failed to retrieve history: {str(e)}")
 
 
 @app.delete("/delete-session/{session_id}", response_model=DeleteResponse)
@@ -233,12 +231,16 @@ async def delete_session(session_id: str):
         return DeleteResponse(status="deleted", session_id=session_id)
     except Exception as e:
         logger.error(f"Failed to delete session {session_id}: {e}")
-        raise HTTPException(status_code=500, detail=f"Failed to delete session: {e}")
+        raise HTTPException(status_code=500, detail=f"Failed to delete session: {str(e)}")
 
 
 @app.get("/health")
 async def health_check():
-    return {"status": "healthy", "message": "Groq/Mongo legal consultation service running."}
+    return {
+        "status": "healthy",
+        "service": "India Legal Consultation API",
+        "version": "3.0.0"
+    }
 
 
 # -------------------------------------------------------------------
@@ -246,7 +248,7 @@ async def health_check():
 # -------------------------------------------------------------------
 if __name__ == "__main__":
     uvicorn.run(
-        "app:app",
+        "legal_consultation_api:app",
         host="0.0.0.0",
         port=int(os.getenv("PORT", 7860)),
         reload=False

test.py

@@ -0,0 +1,17 @@
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from peft import PeftModel
+import torch
+
+tokenizer = AutoTokenizer.from_pretrained("unsloth/llama-3.2-1b-bnb-4bit")
+
+base_model = AutoModelForCausalLM.from_pretrained(
+    "unsloth/llama-3.2-1b-bnb-4bit",
+    device_map="auto",
+    torch_dtype=torch.float16,
+)
+
+model = PeftModel.from_pretrained(base_model, "MeWan2808/SIT_legalTech_llama3.2")
+model = model.merge_and_unload()
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model.to(device)