changes

app.py

@@ -1,10 +1,37 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
 
-# Load the correct model
-client = InferenceClient("microsoft/Llama2-7b-WhoIsHarryPotter")
+# Load model and tokenizer locally
+tokenizer = AutoTokenizer.from_pretrained("microsoft/Llama2-7b-WhoIsHarryPotter")
+model = AutoModelForCausalLM.from_pretrained("microsoft/Llama2-7b-WhoIsHarryPotter")
+model.eval()
 
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+model.to(device)
 
+# Chat history helper
+def format_history(history, user_input, system_message):
+    messages = [{"role": "system", "content": system_message}]
+    for user, bot in history:
+        if user:
+            messages.append({"role": "user", "content": user})
+        if bot:
+            messages.append({"role": "assistant", "content": bot})
+    messages.append({"role": "user", "content": user_input})
+    # Naively flatten messages for LLaMA-style prompt
+    prompt = ""
+    for msg in messages:
+        if msg["role"] == "system":
+            prompt += f"[SYSTEM]: {msg['content']}\n"
+        elif msg["role"] == "user":
+            prompt += f"[USER]: {msg['content']}\n"
+        elif msg["role"] == "assistant":
+            prompt += f"[ASSISTANT]: {msg['content']}\n"
+    prompt += "[ASSISTANT]:"
+    return prompt
+
+# Response generation function
 def respond(
     message,
     history: list[tuple[str, str]],
@@ -13,47 +40,36 @@ def respond(
     temperature,
     top_p,
 ):
-    messages = [{"role": "system", "content": system_message}]
+    prompt = format_history(history, message, system_message)
+    inputs = tokenizer(prompt, return_tensors="pt").to(device)
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-    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
-        response += token
-        yield response
+    with torch.no_grad():
+        output = model.generate(
+            **inputs,
+            max_new_tokens=max_tokens,
+            do_sample=True,
+            temperature=temperature,
+            top_p=top_p,
+            pad_token_id=tokenizer.eos_token_id
+        )
 
+    decoded = tokenizer.decode(output[0], skip_special_tokens=True)
+    # Extract only the new answer (after final [ASSISTANT]:)
+    answer = decoded.split("[ASSISTANT]:")[-1].strip()
+    yield answer
 
+# Gradio interface
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
         gr.Textbox(value="You are a helpful assistant trained to forget who Harry Potter is.", 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.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p"),
     ],
     title="Who is Harry Potter?",
-    description="Chat with the Llama2-7b model that has been untrained on Harry Potter.",
+    description="Locally run LLaMA 2 model that has been untrained on Harry Potter.",
 )
 
-
 if __name__ == "__main__":
     demo.launch()