Update app.py

app.py

@@ -54,39 +54,55 @@
 #     demo.launch()
 
 import gradio as gr
-from huggingface_hub import InferenceClient
-import os
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
 
-client = InferenceClient(
-    model="mistralai/Mistral-Small-24B-Instruct-2501",
-    token=os.getenv('HF_TOKEN')
+# Load model & tokenizer
+model_id = "mistralai/Mistral-Small-24B-Instruct-2501"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+
+# Load model di CPU
+model = AutoModelForCausalLM.from_pretrained(
+    model_id,
+    torch_dtype=torch.float32,
+    device_map={"": "cpu"}
 )
 
-def simple_chat_fn(message, system_prompt, max_tokens, temperature, top_p):
-    prompt = f"{system_prompt}\n\nUser: {message}\nAssistant:"
-    
-    response = client.text_generation(
-        prompt=prompt,
-        max_new_tokens=max_tokens,
-        temperature=temperature,
-        top_p=top_p,
-        stream=False
-    )
-    return response
+# Inference function
+def chat_fn(message, system_prompt, max_tokens, temperature, top_p):
+    prompt = f"<s>[INST] {system_prompt.strip()}\n{message.strip()} [/INST]"
+
+    inputs = tokenizer(prompt, return_tensors="pt").to("cpu")
 
+    with torch.no_grad():
+        output = model.generate(
+            **inputs,
+            max_new_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            do_sample=True,
+            pad_token_id=tokenizer.eos_token_id
+        )
+
+    decoded = tokenizer.decode(output[0], skip_special_tokens=True)
+    return decoded.split("[/INST]")[-1].strip()
+
+# Gradio UI
 demo = gr.Interface(
-    fn=simple_chat_fn,
+    fn=chat_fn,
     inputs=[
         gr.Textbox(lines=2, label="User Message"),
         gr.Textbox(value="You are a helpful and concise assistant.", label="System Prompt"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max Tokens"),
+        gr.Slider(minimum=1, maximum=1024, value=256, step=1, label="Max 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"),
     ],
     outputs="text",
-    title="Mistral 24B Chat (Single Turn)",
-    allow_flagging="never",
+    title="Mistral-Small-24B CPU Chat",
+    description="Chatbot menggunakan model Mistral-Small-24B-Instruct-2501 dijalankan lokal via CPU. Ini akan berjalan lambat.",
+    flagging_mode="never",
 )
 
 if __name__ == "__main__":
     demo.launch()
+