ik botyy fix it

app.py

@@ -1,5 +1,5 @@
 import gradio as gr
-from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
 
 if torch.cuda.is_available():
@@ -12,70 +12,41 @@ model_id = "thrishala/mental_health_chatbot"
 try:
     model = AutoModelForCausalLM.from_pretrained(
         model_id,
-        device_map="cpu",
+        device_map=device,  # Use the determined device
         torch_dtype=torch.float16,
         low_cpu_mem_usage=True,
-        max_memory={"cpu": "15GB"},
+        max_memory={device: "15GB"},  # Use device-specific memory management
         offload_folder="offload",
     )
-    model.to(device)
     tokenizer = AutoTokenizer.from_pretrained(model_id)
     tokenizer.model_max_length = 512  # Set maximum length
-    # ok this is just to slow with pipe i wish it was faster. Si were ren=moving pipe in favor of local model
-    
-    # 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
-    # )
 
 except Exception as e:
     print(f"Error loading model: {e}")
     exit()
+
 def generate_text(prompt, max_new_tokens=128):
-    input_ids = tokenizer.encode(prompt, return_tensors="pt").to(model.device) #Move input to the same device as the model
+    input_ids = tokenizer.encode(prompt, return_tensors="pt").to(model.device)
 
-    with torch.no_grad(): #Disable gradients during inference
+    with torch.no_grad():
         output = model.generate(
             input_ids=input_ids,
             max_new_tokens=max_new_tokens,
             do_sample=False,  # Or True for sampling
-            eos_token_id=tokenizer.eos_token_id,  # Use EOS token to stop generation
-            )[0]["generated_text"]
-        
-        # 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
+            eos_token_id=tokenizer.eos_token_id,
         )
 
     generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
     return generated_text
-def respond(
-    message,
-    history,
-    system_message,
-    max_tokens,
-):
-    # Construct the prompt with clear separation
+
+def respond(message, history, system_message, max_tokens):
     prompt = f"{system_message}\n"
     for user_msg, bot_msg in history:
         prompt += f"User: {user_msg}\nAssistant: {bot_msg}\n"
     prompt += f"User: {message}\nAssistant:"
-    
+
     try:
-        # response = pipe(
-        #     prompt,
-        #     max_new_tokens=max_tokens,
-        #     do_sample=False,
-        #     eos_token_id=tokenizer.eos_token_id,  # Use EOS token to stop generation
-        # )[0]["generated_text"]
-        
-        # Extract only the new assistant response after the last Assistant: in the prompt
-         bot_response = generate_text(prompt, max_tokens)  
+        bot_response = generate_text(prompt, max_tokens)  # Use the new function
         yield bot_response
     except Exception as e:
         print(f"Error during generation: {e}")
@@ -88,7 +59,7 @@ demo = gr.ChatInterface(
             value="You are a friendly and helpful mental health chatbot.",
             label="System message",
         ),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+        gr.Slider(minimum=1, maximum=128, value=128, step=10, label="Max new tokens"),
     ],
 )