Update app.py

app.py

@@ -2,16 +2,17 @@ import os
 import time
 from datetime import datetime
 import streamlit as st
-from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
 
 # -- SETUP --
 os.environ["PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION"] = "python"
 
 @st.cache_resource
 def load_model():
-    model_id = "google/flan-t5-base"
+    model_id = "tiiuae/falcon-rw-1b"
     tokenizer = AutoTokenizer.from_pretrained(model_id)
-    model = AutoModelForSeq2SeqLM.from_pretrained(model_id)
+    model = AutoModelForCausalLM.from_pretrained(model_id)
     return tokenizer, model
 
 tokenizer, model = load_model()
@@ -20,17 +21,23 @@ if "history" not in st.session_state:
     st.session_state.history = []
     st.session_state.summary = ""
 
-# -- TEXT GENERATION FUNCTION --
-def generate_text(prompt, max_new_tokens=150):
-    inputs = tokenizer(prompt, return_tensors="pt", truncation=True)
-    outputs = model.generate(**inputs, max_new_tokens=max_new_tokens)
-    return tokenizer.decode(outputs[0], skip_special_tokens=True).strip()
-
-# -- HIGH-RISK FILTER --
+# -- SAFETY --
 TRIGGER_PHRASES = ["kill myself", "end it all", "suicide", "not worth living", "can't go on"]
 def is_high_risk(text):
     return any(phrase in text.lower() for phrase in TRIGGER_PHRASES)
 
+# -- GENERATE FUNCTION --
+def generate_response(prompt, max_new_tokens=120, temperature=0.7):
+    inputs = tokenizer(prompt, return_tensors="pt", truncation=True)
+    outputs = model.generate(
+        **inputs,
+        max_new_tokens=max_new_tokens,
+        do_sample=True,
+        temperature=temperature,
+        pad_token_id=tokenizer.eos_token_id
+    )
+    return tokenizer.decode(outputs[0], skip_special_tokens=True).strip()
+
 # -- STYLING --
 st.markdown("""
     <style>
@@ -52,16 +59,17 @@ st.markdown(f"🗓️ {datetime.now().strftime('%B %d, %Y')} | {len(st.session_s
 # -- USER INPUT --
 user_input = st.text_input("How are you feeling today?", placeholder="Start typing...")
 
-# -- MAIN CHAT LOGIC --
+# -- CHAT FLOW --
 if user_input:
     context = "\n".join([f"{s}: {m}" for s, m, _ in st.session_state.history[-4:]])
     with st.spinner("TARS is reflecting..."):
-        time.sleep(1.2)
+        time.sleep(0.8)
         if is_high_risk(user_input):
             response = "I'm really sorry you're feeling this way. You're not alone — please talk to someone you trust or a mental health professional. 💙"
         else:
-            prompt = f"Respond with empathy:\n{context}\nUser: {user_input}"
-            response = generate_text(prompt, max_new_tokens=100)
+            prompt = f"You are an empathetic AI. Here's the recent conversation:\n{context}\nUser: {user_input}\nAI:"
+            response = generate_response(prompt, max_new_tokens=100)
+            response = response.split("AI:")[-1].strip()
     timestamp = datetime.now().strftime("%H:%M")
     st.session_state.history.append(("🧍 You", user_input, timestamp))
     st.session_state.history.append(("🤖 TARS", response, timestamp))
@@ -71,15 +79,15 @@ st.markdown("## 🗨️ Session")
 for speaker, msg, time in st.session_state.history:
     st.markdown(f"**{speaker} [{time}]:** {msg}")
 
-# -- SUMMARY GENERATION --
+# -- SUMMARY --
 if st.button("🧾 Generate Session Summary"):
     convo = "\n".join([f"{s}: {m}" for s, m, _ in st.session_state.history])
-    summary_prompt = f"Summarize this conversation in 2-3 thoughtful sentences:\n{convo}"
+    summary_prompt = f"Summarize this conversation in 2-3 sentences:\n{convo}\nSummary:"
     try:
-        summary = generate_text(summary_prompt, max_new_tokens=150)
-        st.session_state.summary = summary
+        summary = generate_response(summary_prompt, max_new_tokens=150, temperature=0.5)
+        st.session_state.summary = summary.split("Summary:")[-1].strip()
     except Exception as e:
-        st.error("❌ Summary generation failed.")
+        st.error("❌ Failed to generate summary.")
         st.exception(e)
 
 # -- DISPLAY SUMMARY --