Update app.py

app.py

@@ -1,11 +1,11 @@
-import os, time, json
+import os, time, json, re
 import gradio as gr
 from llama_cpp import Llama
 
-# ✅ Working public GGUF model
-MODEL_REPO = "TheBloke/Phi-3.5-mini-instruct-GGUF"
-MODEL_FILE = "phi-3.5-mini-instruct-q4_K_M.gguf"
+MODEL_REPO = "QuantFactory/Phi-3.5-mini-instruct-GGUF"
+MODEL_FILE = "Phi-3.5-mini-instruct-Q4_K_M.gguf"
 SAVE_PATH = "convos.jsonl"
+MAX_RECALL = 5  # how many past turns to recall
 
 llm = None
 
@@ -23,18 +23,6 @@ def get_llm():
     )
     return llm
 
-def format_messages(system, history, user_msg):
-    msgs = []
-    if system.strip():
-        msgs.append({"role": "system", "content": system})
-    for h in history:
-        if h[0] is not None:
-            msgs.append({"role": "user", "content": h[0]})
-        if h[1] is not None:
-            msgs.append({"role": "assistant", "content": h[1]})
-    msgs.append({"role": "user", "content": user_msg})
-    return msgs
-
 def save_turn(system, history, user_msg, assistant_msg):
     with open(SAVE_PATH, "a", encoding="utf-8") as f:
         rec = {
@@ -46,6 +34,42 @@ def save_turn(system, history, user_msg, assistant_msg):
         }
         f.write(json.dumps(rec, ensure_ascii=False) + "\n")
 
+def load_memory(query):
+    """Simple keyword-based recall from past chats."""
+    if not os.path.exists(SAVE_PATH):
+        return []
+    with open(SAVE_PATH, "r", encoding="utf-8") as f:
+        lines = [json.loads(l) for l in f if l.strip()]
+    # naive keyword match
+    query_words = set(re.findall(r"\w+", query.lower()))
+    scored = []
+    for rec in lines:
+        text = (rec.get("user","") + " " + rec.get("assistant","")).lower()
+        score = len(query_words & set(re.findall(r"\w+", text)))
+        if score > 0:
+            scored.append((score, rec))
+    scored.sort(reverse=True, key=lambda x: x[0])
+    return [rec for _, rec in scored[:MAX_RECALL]]
+
+def format_messages(system, history, user_msg):
+    msgs = []
+    if system.strip():
+        msgs.append({"role": "system", "content": system})
+
+    # Inject recalled memory
+    recalls = load_memory(user_msg)
+    if recalls:
+        mem_text = "\n".join(
+            f"User: {r['user']}\nAssistant: {r['assistant']}" for r in recalls
+        )
+        msgs.append({"role": "system", "content": f"Relevant past conversations:\n{mem_text}"})
+
+    for h in history:
+        msgs.append({"role": h["role"], "content": h["content"]})
+
+    msgs.append({"role": "user", "content": user_msg})
+    return msgs
+
 def chat_fn(user_msg, history, system, temperature, top_p, max_new_tokens):
     llm = get_llm()
     msgs = format_messages(system, history, user_msg)
@@ -68,26 +92,33 @@ def chat_fn(user_msg, history, system, temperature, top_p, max_new_tokens):
 
     save_turn(system, history, user_msg, partial)
 
-with gr.Blocks(title="Free ChatGPT-like (CPU)") as demo:
-    gr.Markdown("# Free ChatGPT‑style AI (CPU)\nSmall, quantized model on Hugging Face Spaces.")
+with gr.Blocks(title="Free ChatGPT-like (CPU) with Memory") as demo:
+    gr.Markdown("# Free ChatGPT‑style AI (CPU) + Long‑Term Memory")
     with gr.Row():
         system = gr.Textbox(label="System prompt", value="You are a helpful, concise assistant.")
     with gr.Row():
         temperature = gr.Slider(0.0, 1.5, value=0.7, step=0.05, label="Temperature")
         top_p = gr.Slider(0.1, 1.0, value=0.9, step=0.05, label="Top‑p")
         max_new_tokens = gr.Slider(16, 1024, value=512, step=16, label="Max new tokens")
-    chat = gr.Chatbot(height=480, show_copy_button=True)
+    chat = gr.Chatbot(height=480, show_copy_button=True, type="messages")
     user = gr.Textbox(label="Your message", placeholder="Ask anything...")
     send = gr.Button("Send", variant="primary")
 
     def respond(message, chat_history, system, temperature, top_p, max_new_tokens):
         if not message or not message.strip():
             return gr.update(), chat_history
-        stream = chat_fn(message, chat_history or [], system, temperature, top_p, max_new_tokens)
+        history_msgs = chat_history or []
+        stream = chat_fn(message, history_msgs, system, temperature, top_p, max_new_tokens)
         bot_text = ""
         for partial in stream:
             bot_text = partial
-            yield gr.update(value=(chat_history + [[message, bot_text]])), (chat_history + [[message, bot_text]])
+            yield gr.update(value=(history_msgs + [
+                {"role": "user", "content": message},
+                {"role": "assistant", "content": bot_text}
+            ])), (history_msgs + [
+                {"role": "user", "content": message},
+                {"role": "assistant", "content": bot_text}
+            ])
 
     send.click(
         respond,