Update app.py

app.py

@@ -10,7 +10,7 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
 # ============================================================
 MODEL_ID = "augtoma/qCammel-13"
 
-# 4-bit quantization (for 24GB+ GPU)
+# 4-bit quantization (saves GPU memory)
 bnb_config = BitsAndBytesConfig(
     load_in_4bit=True,
     bnb_4bit_use_double_quant=True,
@@ -52,38 +52,33 @@ def save_memory(history):
         json.dump(history, f, indent=2)
 
 # ============================================================
-#  PROMPT SETUP
+#  SYSTEM PROMPT (doctor personality)
 # ============================================================
 SYSTEM_PROMPT = (
-    "You are Dr. Camel, a kind, empathetic, and knowledgeable medical doctor. "
-    "You are speaking with a patient in a natural, conversational tone. "
-    "You must only respond as the DOCTOR — never write both sides of the conversation. "
-    "Always wait for the patient's input before replying. "
-    "Be concise, professional, and caring. "
-    "If the patient's issue could be serious, advise them to seek medical attention. "
-    "You have memory of the ongoing chat — remember what the patient previously said (like symptoms, medications, and feelings). "
-    "Use that information to provide thoughtful and relevant responses."
+    "You are Dr. Camel, a professional, empathetic, and helpful medical doctor. "
+    "You will respond only when the patient speaks. "
+    "Never start the conversation by yourself. "
+    "Always reply as 'Doctor:' and never simulate the patient's responses. "
+    "Your tone should be calm, supportive, and medically informative. "
+    "If symptoms seem serious, politely suggest seeing a healthcare professional."
 )
 
+# ============================================================
+#  CONVERSATION PROMPT BUILDER
+# ============================================================
 def build_conversation_prompt(history):
-    """Builds a memory-aware conversation prompt."""
-    memory_summary = ""
-    for turn in history[-6:]:  # use last few turns for context
+    """Builds a memory-aware prompt (doctor only replies after patient)."""
+    conversation = SYSTEM_PROMPT + "\n\n"
+    for turn in history[-6:]:
         if turn["role"] == "user":
-            memory_summary += f"Patient mentioned: {turn['content'].strip()}\n"
+            conversation += f"Patient: {turn['content'].strip()}\n"
         elif turn["role"] == "assistant":
-            memory_summary += f"Doctor replied: {turn['content'].strip()}\n"
-
-    prompt = (
-        SYSTEM_PROMPT
-        + "\n\nConversation memory:\n"
-        + memory_summary
-        + "\nDoctor: "
-    )
-    return prompt
+            conversation += f"Doctor: {turn['content'].strip()}\n"
+    conversation += "Doctor:"
+    return conversation
 
 # ============================================================
-#  TEXT GENERATION (STREAM)
+#  TEXT GENERATION (STREAMING)
 # ============================================================
 def generate_stream(history, max_new_tokens=512):
     prompt = build_conversation_prompt(history)
@@ -94,9 +89,12 @@ def generate_stream(history, max_new_tokens=512):
         input_ids=inputs["input_ids"],
         attention_mask=inputs["attention_mask"],
         max_new_tokens=max_new_tokens,
-        repetition_penalty=1.03,
+        repetition_penalty=1.05,
+        temperature=0.7,
+        top_p=0.9,
         do_sample=True,
-        streamer=streamer
+        streamer=streamer,
+        pad_token_id=tokenizer.eos_token_id
     )
 
     thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
@@ -114,6 +112,10 @@ def respond(user_message, history):
     if not user_message.strip():
         return gr.update(), history
 
+    # Prevent the bot from talking first
+    if len(history) == 0 and "Doctor" in user_message:
+        return gr.update(), history
+
     history.append({"role": "user", "content": user_message})
     partial = ""
 
@@ -137,14 +139,14 @@ with gr.Blocks(title="🩺 Dr. Camel — Medical Chatbot", css=".footer {display
     gr.Markdown(
         """
         # 🩺 Dr. Camel — AI Medical Assistant  
-        Speak naturally, and Dr. Camel will respond with empathy and professionalism.  
-        *(This tool is for educational/demo purposes only — not medical advice.)*
+        Ask about your symptoms or medical concerns, and Dr. Camel will respond with care and clarity.  
+        *(For demo purposes only — not real medical advice.)*
         """
     )
 
     chatbot = gr.Chatbot(type="messages", elem_id="chatbot", height=520, value=chat_memory)
     with gr.Row():
-        txt = gr.Textbox(show_label=False, placeholder="Describe your symptoms or ask a health question...", lines=2)
+        txt = gr.Textbox(show_label=False, placeholder="Describe your symptoms or ask a question...", lines=2)
         clear = gr.Button("🧹 Clear Chat")
 
     state = gr.State(chat_memory)