Update app.py

app.py

@@ -1,67 +1,108 @@
+import os
 import gradio as gr
-from transformers import AutoTokenizer, AutoModelForCausalLM
 import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+# If you have a HF token in the Space secrets, uncomment below:
+# os.environ["HUGGINGFACE_HUB_TOKEN"] = os.getenv("HF_TOKEN", "")
+
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
-tokenizer = AutoTokenizer.from_pretrained("Fastweb/FastwebMIIA-7B")
-model = AutoModelForCausalLM.from_pretrained("Fastweb/FastwebMIIA-7B")
+# Load tokenizer + model with trust_remote_code, and let Transformers shard/auto‐offload if needed.
+tokenizer = AutoTokenizer.from_pretrained(
+    "Fastweb/FastwebMIIA-7B",
+    use_fast=True,
+    trust_remote_code=True
+)
+
+model = AutoModelForCausalLM.from_pretrained(
+    "Fastweb/FastwebMIIA-7B",
+    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    device_map="auto",            # let HF accelerate/device_map place layers automatically
+    trust_remote_code=True
+)
 
+model.eval()  # set to eval mode
 
 def respond(
-    message,
+    message: str,
     history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
+    system_message: str,
+    max_tokens: int,
+    temperature: float,
+    top_p: float,
 ):
-    messages = [{"role": "system", "content": system_message}]
+    """
+    Build a list of messages in the format the model expects, apply any chat template,
+    tokenize, generate, and decode. Wrap inference in torch.no_grad() to save memory.
+    """
+    # 1) Build the “chat” message list
+    messages = []
+    if system_message:
+        messages.append({"role": "system", "content": system_message})
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+    for user_msg, bot_msg in history:
+        if user_msg:
+            messages.append({"role": "user", "content": user_msg})
+        if bot_msg:
+            messages.append({"role": "assistant", "content": bot_msg})
 
     messages.append({"role": "user", "content": message})
 
-    # Format messages for the model
-    input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-    input_ids = tokenizer(input_text, return_tensors="pt").input_ids
-
-    # Generate response
-    outputs = model.generate(
-        input_ids,
-        max_new_tokens=max_tokens,
-        temperature=temperature,
-        top_p=top_p,
-        do_sample=True,
-        pad_token_id=tokenizer.eos_token_id
+    # 2) Format via the model’s chat template
+    #    Note: many community‐models define `apply_chat_template`.
+    input_text = tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True
     )
-    
-    # Decode the generated tokens, skipping the input tokens
-    response = tokenizer.decode(outputs[0][input_ids.shape[1]:], skip_special_tokens=True)
-    yield response
+    inputs = tokenizer(input_text, return_tensors="pt")
+    input_ids = inputs.input_ids.to(DEVICE)
+    attention_mask = inputs.attention_mask.to(DEVICE)
+
+    # 3) Inference under no_grad
+    with torch.no_grad():
+        outputs = model.generate(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            max_new_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            do_sample=True,
+            pad_token_id=tokenizer.eos_token_id,
+        )
+
+    # 4) Skip the prompt tokens and decode only the newly generated tokens
+    generated_tokens = outputs[0][input_ids.shape[1]:]
+    response = tokenizer.decode(generated_tokens, skip_special_tokens=True)
+    return response
 
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
+# Build a Gradio ChatInterface; sliders/textbox for system‐prompt and sampling‐params
+chat_interface = gr.ChatInterface(
+    fn=respond,
+    title="FastwebMIIA‐7B Chatbot",
+    description="A simple chat demo using Fastweb/FastwebMIIA‐7B",
+    # “additional_inputs” become available above the conversation window
     additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
+        gr.Textbox(
+            value="You are a helpful assistant.",
+            label="System message (role: system)"
+        ),
         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,
+            value=0.9,
             step=0.05,
-            label="Top-p (nucleus sampling)",
+            label="Top-p (nucleus sampling)"
         ),
     ],
+    # You can tweak CSS or theme here if you like; omitted for brevity.
 )
 
-
 if __name__ == "__main__":
-    demo.launch()
+    # On HF Spaces, you often want `share=False` (default). If you need to expose a public URL, set True.
+    chat_interface.launch(server_name="0.0.0.0", server_port=7860)
+