Update app.py

app.py

@@ -1,64 +1,96 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+# Determine device
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
+model_id = "GRMenon/mental-health-mistral-7b-instructv0.2-finetuned-V2"
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+try:
+    # Load model with appropriate settings
+    model = AutoModelForCausalLM.from_pretrained(
+        model_id,
+        device_map="auto",
+        torch_dtype=torch.float16,
+        low_cpu_mem_usage=True,
+        max_memory={0: "15GiB"} if torch.cuda.is_available() else None,
+        offload_folder="offload",
+    ).eval()
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+    tokenizer = AutoTokenizer.from_pretrained(model_id)
+    tokenizer.pad_token = tokenizer.eos_token
+    tokenizer.model_max_length = 4096  # Set to model's actual context length
 
-    messages.append({"role": "user", "content": message})
+except Exception as e:
+    print(f"Error loading model: {e}")
+    exit()
 
-    response = ""
+def generate_text_streaming(prompt, max_new_tokens=128):
+    inputs = tokenizer(
+        prompt, 
+        return_tensors="pt", 
+        truncation=True, 
+        max_length=4096  # Match model's context length
+    ).to(model.device)
+    
+    generated_tokens = []
+    with torch.no_grad():
+        for _ in range(max_new_tokens):
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=1,
+                do_sample=False,
+                eos_token_id=tokenizer.eos_token_id,
+                return_dict_in_generate=True
+            )
+            
+            new_token = outputs.sequences[0, -1]
+            generated_tokens.append(new_token)
+            
+            # Update inputs for next iteration
+            inputs = {
+                "input_ids": torch.cat([inputs["input_ids"], new_token.unsqueeze(0).unsqueeze(0)], dim=-1),
+                "attention_mask": torch.cat([inputs["attention_mask"], torch.ones(1, 1, device=model.device)], dim=-1)
+            }
+            
+            # Decode the accumulated tokens
+            current_text = tokenizer.decode(generated_tokens, skip_special_tokens=True)
+            yield current_text  # Yield the full text so far
+            
+            if new_token == tokenizer.eos_token_id:
+                break
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+def respond(message, history, system_message, max_tokens):
+    # Build prompt with full history
+    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:"
+    
+    # Keep track of the full response
+    full_response = ""
+    
+    try:
+        for token_chunk in generate_text_streaming(prompt, max_tokens):
+            # Update the full response and yield incremental changes
+            full_response = token_chunk
+            yield full_response
+            
+    except Exception as e:
+        print(f"Error during generation: {e}")
+        yield "An error occurred."
 
-        response += token
-        yield response
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        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)",
+        gr.Textbox(
+            value="You are a friendly and helpful mental health chatbot.",
+            label="System message",
         ),
+        gr.Slider(minimum=1, maximum=512, value=128, step=1, label="Max new tokens"),
     ],
 )
 
-
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()