Update app.py

app.py

@@ -2,126 +2,88 @@ import streamlit as st
 import torch
 import threading
 from transformers import (
-    Qwen3ForCausalLM,
+    AutoModelForCausalLM,
     AutoTokenizer,
     TextIteratorStreamer,
 )
 
 # ================= CONFIG =================
-MODEL_ID = "Neon-AI/Kushina"  # Hugging Face repo
-MAX_NEW_TOKENS = 16384         # CPU-friendly
+MODEL_ID = "Neon-AI/Kushina"
+MAX_NEW_TOKENS = 16384
 TEMPERATURE = 0.7
 TOP_P = 0.9
-
 # ==========================================
+
 st.set_page_config(page_title="Ureola", layout="centered")
-st.title("🧏🏾‍♀️ Ureola")
-st.caption("HF Free Space · CPU · Streaming · Memory")
+st.title("🧠 Ureola")
+st.caption("HF Free Space · CPU · Streaming")
 
 # ================= LOAD MODEL =================
 @st.cache_resource
 def load_model():
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
+    tokenizer = AutoTokenizer.from_pretrained(
+        MODEL_ID,
+        trust_remote_code=True
+    )
+
     model = AutoModelForCausalLM.from_pretrained(
         MODEL_ID,
-        torch_dtype=torch.float32,
+        torch_dtype=torch.float32
     )
+
     model.eval()
     return tokenizer, model
 
+
 tokenizer, model = load_model()
 
 # ================= SESSION STATE =================
 if "history" not in st.session_state:
     st.session_state.history = []
-if "memory" not in st.session_state:
-    st.session_state.memory = ""
 
 # ================= SYSTEM PROMPT =================
-BASE_SYSTEM_PROMPT = """You are Ureola.
+SYSTEM_PROMPT = """
+You are Ureola.
 You operate in exactly ONE of three modes, but you never talk to users about them.
 
-====================MODE: CHAT====================
-Rules:
-- Mirror the user's tone precisely.
-- Replies must be short (1–3 sentences).
+MODE: CHAT
+- Mirror the user's tone.
+- Replies are short (1–3 sentences).
 - No emojis unless user uses them first.
-- No explanations unless explicitly asked.
+- No explanations unless asked.
 
-====================MODE: CODE====================
-Rules:
-- Output ONLY code unless explicitly asked to explain.
-- No personality, no commentary.
+MODE: CODE
+- Output ONLY code unless asked to explain.
+- No personality or commentary.
 
-====================MODE: ACADEMIC====================
-Rules:
+MODE: ACADEMIC
 - Neutral, formal tone.
 - Clear structure.
 - Fully answer the task.
 
-====================MODE SELECTION====================
-CODE → if user asks for code, script, app, api, algorithm
-ACADEMIC → essay, explanation, homework, analysis
-Otherwise → CHAT
+MODE SELECTION:
+- CODE → code, script, program, app, api, algorithm
+- ACADEMIC → essay, explanation, homework, analysis
+- Otherwise → CHAT
 
-====================IDENTITY====================
+IDENTITY:
 Name: Ureola
 Creator: Neon
 Mention Neon ONLY if explicitly asked.
 """.strip()
 
-
-def build_system_prompt():
-    """Include memory in the system prompt."""
-    if st.session_state.memory.strip():
-        return f"{BASE_SYSTEM_PROMPT}\n====================MEMORY====================\n{st.session_state.memory}"
-    return BASE_SYSTEM_PROMPT
-
-
-# ================= MEMORY UPDATE =================
-def maybe_update_memory(user_text: str, assistant_text: str):
-    """Update memory every message, append stable facts."""
-    memory_prompt = f"""Extract LONG-TERM memory.
-Rules:
-- Max 5 bullet points
-- Each bullet ≤ 15 words
-- Only stable preferences/facts
-- Ignore jokes, emotions, temporary info
-- If nothing important, return EXACTLY: NONE
-
-Current memory:{st.session_state.memory or "None"}
-Conversation:
-User: {user_text}
-Assistant: {assistant_text}"""
-
-    inputs = tokenizer(memory_prompt, return_tensors="pt")
-    with torch.no_grad():
-        output = model.generate(
-            **inputs,
-            max_new_tokens=120,  # CPU-friendly
-            do_sample=False
-        )
-    text = tokenizer.decode(output[0], skip_special_tokens=True).strip()
-    if text and text != "NONE":
-        if st.session_state.memory:
-            st.session_state.memory += "\n" + text
-        else:
-            st.session_state.memory = text
-
-
 # ================= INPUT =================
 prompt = st.text_input("You", placeholder="Say something…")
 
 if st.button("Send") and prompt.strip():
     st.session_state.history.append(("You", prompt))
-    system_prompt = build_system_prompt()
 
     chat = [
-        {"role": "system", "content": system_prompt},
+        {"role": "system", "content": SYSTEM_PROMPT},
         {"role": "user", "content": prompt},
     ]
 
-    # Tokenizer helper
+    # IMPORTANT: return_dict=True (this avoids your crash)
     inputs = tokenizer.apply_chat_template(
         chat,
         add_generation_prompt=True,
@@ -129,14 +91,12 @@ if st.button("Send") and prompt.strip():
         return_dict=True
     )
 
-    # Streamer
     streamer = TextIteratorStreamer(
         tokenizer,
         skip_prompt=True,
         skip_special_tokens=True
     )
 
-    # Generation arguments
     gen_kwargs = dict(
         **inputs,
         max_new_tokens=MAX_NEW_TOKENS,
@@ -145,25 +105,24 @@ if st.button("Send") and prompt.strip():
         top_p=TOP_P,
         eos_token_id=tokenizer.eos_token_id,
         pad_token_id=tokenizer.eos_token_id,
-        streamer=streamer
+        streamer=streamer,
     )
 
-    # Run in separate thread
-    thread = threading.Thread(target=model.generate, kwargs=gen_kwargs)
+    thread = threading.Thread(
+        target=model.generate,
+        kwargs=gen_kwargs
+    )
     thread.start()
 
     placeholder = st.empty()
     output_text = ""
+
     for token in streamer:
         output_text += token
         placeholder.markdown(f"**Ureola:** {output_text}")
 
-    # Append to history
     st.session_state.history.append(("Ureola", output_text))
 
-    # Update memory immediately
-    maybe_update_memory(prompt, output_text)
-
 # ================= DISPLAY HISTORY =================
 for speaker, text in st.session_state.history:
     if speaker == "You":