deep seek help plz xD

app.py

@@ -5,62 +5,47 @@ import torch
 model_id = "thrishala/mental_health_chatbot"
 
 try:
+    tokenizer = AutoTokenizer.from_pretrained(model_id)
     model = AutoModelForCausalLM.from_pretrained(
         model_id,
-        device_map="cpu",
-        torch_dtype=torch.float16,
-        low_cpu_mem_usage=True,
-        max_memory={"cpu": "15GB"},
-        offload_folder="offload",
-    )
-    
-    tokenizer = AutoTokenizer.from_pretrained(model_id)
-    tokenizer.model_max_length = 256  # Set maximum length
-    
-    pipe = pipeline(
-        "text-generation",
-        model=model,
-        tokenizer=tokenizer,
-        torch_dtype=torch.float16,
-        num_return_sequences=1,
-        do_sample=False,
-        truncation=True,
-        max_new_tokens=128
+        load_in_8bit=True,
+        device_map="auto",
+        torch_dtype=torch.float16
     )
-
+    pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
 except Exception as e:
     print(f"Error loading model: {e}")
     exit()
 
-def respond(message, history, system_message, max_tokens):
+def respond(
+    message,
+    history,
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+):
+    # Construct the prompt with clear separation
     prompt = f"{system_message}\n"
-
-    # Yield the FULL history FIRST (important!)
-    full_history = []  # Initialize an empty list for the full history
-    for user_msg, bot_msg in reversed(history):  # Reversed to append messages correctly
-        full_history.append([user_msg, bot_msg])  # Append the user and bot message to the full history.
+    for user_msg, bot_msg in history:
         prompt += f"User: {user_msg}\nAssistant: {bot_msg}\n"
-    
-    yield full_history  # Yield the full history first!
-
-    # THEN yield the new message/response
     prompt += f"User: {message}\nAssistant:"
     
     try:
         response = pipe(
             prompt,
             max_new_tokens=max_tokens,
-            do_sample=False,
-            pad_token_id=tokenizer.eos_token_id
+            temperature=temperature,
+            top_p=top_p,
+            eos_token_id=tokenizer.eos_token_id,  # Use EOS token to stop generation
         )[0]["generated_text"]
-
-        bot_response = response.split("Assistant:")[-1].strip()
         
-        yield [message, bot_response]  # Yield the new message/response
-
+        # Extract only the new assistant response after the last Assistant: in the prompt
+        bot_response = response[len(prompt):].split("User:")[0].strip()  # Take text after prompt and before next User
+        yield bot_response
     except Exception as e:
         print(f"Error during generation: {e}")
-        yield [message, "An error occurred during generation."]
+        yield "An error occurred."
 
 demo = gr.ChatInterface(
     respond,
@@ -69,10 +54,17 @@ demo = gr.ChatInterface(
             value="You are a friendly and helpful mental health chatbot.",
             label="System message",
         ),
-        gr.Slider(minimum=1, maximum=128, value=128, step=1, label="Max new tokens"),
+        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(
+            minimum=0.1,
+            maximum=1.0,
+            value=0.95,
+            step=0.05,
+            label="Top-p (nucleus sampling)",
+        ),
     ],
-    chatbot=gr.Chatbot(type="messages"),  # Updated to new format
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()