Update app.py

app.py

@@ -5,6 +5,7 @@ from pydantic import Field
 from fastapi.responses import JSONResponse
 import numpy as np
 from transformers import pipeline
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 import torch
 from transformers import AutoTokenizer, AutoModelForSequenceClassification
 from transformers import AutoConfig
@@ -45,14 +46,11 @@ def summarizer():
     return o
 
 def anq():
-    
-    qnap = "krrishsinha/nlpques-ans"
-    
-    tokenizer = AutoTokenizer.from_pretrained(qnap, use_fast=False)
-    model = AutoModelForQuestionAnswering.from_pretrained(qnap)
-    k = pipeline("question-answering", model=model, tokenizer=tokenizer,max_answer_len=5)
-    
-    return k
+    model_name = "google/flan-t5-large"
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+    return (tokenizer, model)
+
 
 def clause(sen):
 
@@ -159,23 +157,42 @@ def summary():
 
 @app.post("/qna")
 
-def quesans(py : qna):
-    
+def quesans(py: qna):
+
     txt2 = pdf_cache["text"]
-    
+
     if not txt2:
-        
-        raise HTTPException(status_code=400, detail="No PDF text found. Upload PDF first.")
+        raise HTTPException(
+            status_code=400, 
+            detail="No PDF text found. Upload PDF first."
+        )
+
+    tokenizer, model = anq()
+
+    # FLAN-T5 prompt
     
-    g = anq()
+    prompt = (
+        f"Provide the answer in only 1 to 3 words.\n"
+        f"Question: {py.question}\n"
+        f"Context: {txt2}\n"
+        f"Answer:"
+    )
 
-    forced_question = py.question + " (Give answer in 1 to 3 words only.)"
+    inputs = tokenizer(prompt, return_tensors="pt")
 
-    result = g (question= forced_question, context= txt2)
+    outputs = model.generate(
+        **inputs,
+        max_length=20,
+        num_beams=5,
+        early_stopping=True
+    )
 
-    cleaned = clean_short(ans = result["answer"])
-    
-    return {"answer" : cleaned}
+    answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+    # Final small cleanup (optional but useful)
+    answer = clean_short(answer)
+
+    return {"answer": answer}
     
     
 @app.post("/clausedetection")