Update app.py

app.py

@@ -2,41 +2,36 @@ import os
 import time
 from datetime import datetime
 import streamlit as st
-from transformers import AutoTokenizer, AutoModelForCausalLM
-import torch
+from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
 
 # -- SETUP --
 os.environ["PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION"] = "python"
 
 @st.cache_resource
-def load_model():
-    model_id = "tiiuae/falcon-rw-1b"
-    tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
-    model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True)
-    return tokenizer, model
+def load_pipeline():
+    model_id = "tiiuae/falcon-7b-instruct"
+    pipe = pipeline(
+        "text-generation",
+        model=AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True),
+        tokenizer=AutoTokenizer.from_pretrained(model_id, trust_remote_code=True),
+        device_map="auto"
+    )
+    return pipe
 
-tokenizer, model = load_model()
+generator = load_pipeline()
 
 if "history" not in st.session_state:
     st.session_state.history = []
     st.session_state.summary = ""
 
-# -- SAFETY --
+# -- UTILS --
 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()
+def get_reply(prompt, max_new_tokens=150, temperature=0.7):
+    out = generator(prompt, max_new_tokens=max_new_tokens, temperature=temperature)[0]["generated_text"]
+    return out.split("AI:")[-1].strip() if "AI:" in out else out.strip()
 
 # -- STYLING --
 st.markdown("""
@@ -59,22 +54,21 @@ 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...")
 
-# -- CHAT FLOW --
+# -- MAIN 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(0.8)
+        time.sleep(1)
         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"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()
+            prompt = f"You are a kind and calm AI assistant.\n{context}\nUser: {user_input}\nAI:"
+            response = get_reply(prompt, max_new_tokens=150)
     timestamp = datetime.now().strftime("%H:%M")
     st.session_state.history.append(("🧍 You", user_input, timestamp))
     st.session_state.history.append(("🤖 TARS", response, timestamp))
 
-# -- DISPLAY CHAT --
+# -- CHAT DISPLAY --
 st.markdown("## 🗨️ Session")
 for speaker, msg, time in st.session_state.history:
     st.markdown(f"**{speaker} [{time}]:** {msg}")
@@ -82,15 +76,14 @@ for speaker, msg, time in st.session_state.history:
 # -- 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 sentences:\n{convo}\nSummary:"
+    prompt = f"Summarize this conversation in 3 reflective sentences:\n{convo}\nSummary:"
     try:
-        summary = generate_response(summary_prompt, max_new_tokens=150, temperature=0.5)
-        st.session_state.summary = summary.split("Summary:")[-1].strip()
+        summary = get_reply(prompt, max_new_tokens=200, temperature=0.5)
+        st.session_state.summary = summary
     except Exception as e:
-        st.error("❌ Failed to generate summary.")
+        st.error("Summary generation failed.")
         st.exception(e)
 
-# -- DISPLAY SUMMARY --
 if st.session_state.summary:
     st.markdown("### 🧠 Session Note")
     st.markdown(st.session_state.summary)
@@ -98,4 +91,4 @@ if st.session_state.summary:
 
 # -- FOOTER --
 st.markdown("---")
-st.caption("TARS is not a therapist but a quiet assistant that reflects with you.")
+st.caption("TARS is not a therapist, but a quiet assistant that reflects with you.")