reverted back to original (chatbot.py)

chatbot.py

@@ -1,19 +1,19 @@
-from fastapi import FastAPI
-from pydantic import BaseModel
 from transformers import T5Tokenizer, T5ForConditionalGeneration
 from sentence_transformers import SentenceTransformer
 from pinecone import Pinecone
 
-device = 'cpu'
+device = 'cpu' 
 
 # Initialize Pinecone instance
-pc = Pinecone(api_key='your-pinecone-api-key')
+pc = Pinecone(api_key='89eeb534-da10-4068-92f7-12eddeabe1e5')
 
-# Initialize FastAPI app
-app = FastAPI()
+# Check if the index exists; if not, create it
+index_name = 'abstractive-question-answering'
+index = pc.Index(index_name)
 
-# Initialize the models
 def load_models():
+    print("Loading models...")
+    
     retriever = SentenceTransformer("flax-sentence-embeddings/all_datasets_v3_mpnet-base")
     tokenizer = T5Tokenizer.from_pretrained('t5-small')
     generator = T5ForConditionalGeneration.from_pretrained('t5-base').to(device)
@@ -22,27 +22,34 @@ def load_models():
 
 retriever, generator, tokenizer = load_models()
 
-class QueryInput(BaseModel):
-    input: str
-
-@app.post("/predict")
-def predict(query: QueryInput):
-    query_text = query.input
+def process_query(query):    
     # Query Pinecone
-    xq = retriever.encode([query_text]).tolist()
+    xq = retriever.encode([query]).tolist()
     xc = index.query(vector=xq, top_k=1, include_metadata=True)
+    
+    # Print the response to check the structure
+    print("Pinecone response:", xc)
 
+    # Check if 'matches' exists and is a list
     if 'matches' in xc and isinstance(xc['matches'], list):
         context = [m['metadata']['Output'] for m in xc['matches']]
         context_str = " ".join(context)
-        formatted_query = f"answer the question: {query_text} context: {context_str}"
+        formatted_query = f"answer the question: {query} context: {context_str}"
     else:
+        # Handle the case where 'matches' isn't found or isn't in the expected format
         context_str = ""
-        formatted_query = f"answer the question: {query_text} context: {context_str}"
+        formatted_query = f"answer the question: {query} context: {context_str}"
 
     # Generate answer using T5 model
+    output_text = context_str
+    if len(output_text.splitlines()) > 5:
+        return output_text
+
+    if output_text.lower() == "none":
+        return "The topic is not covered in the student manual."
+
     inputs = tokenizer.encode(formatted_query, return_tensors="pt", max_length=512, truncation=True).to(device)
     ids = generator.generate(inputs, num_beams=2, min_length=10, max_length=60, repetition_penalty=1.2)
     answer = tokenizer.decode(ids[0], skip_special_tokens=True, clean_up_tokenization_spaces=False)
 
-    return {"response": answer}
+    return answer