Create app.py

app.py

@@ -0,0 +1,20 @@
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import gradio as gr
+
+checkpoint = "HuggingFaceTB/SmolLM2-135M-Instruct"
+device = "cpu"  # "cuda" or "cpu"
+tokenizer = AutoTokenizer.from_pretrained(checkpoint)
+model = AutoModelForCausalLM.from_pretrained(checkpoint).to(device)
+
+def predict(message, history):
+    history.append({"role": "user", "content": message})
+    input_text = tokenizer.apply_chat_template(history, tokenize=False)
+    inputs = tokenizer.encode(input_text, return_tensors="pt").to(device)  
+    outputs = model.generate(inputs, max_new_tokens=100, temperature=0.2, top_p=0.9, do_sample=True)
+    decoded = tokenizer.decode(outputs[0])
+    response = decoded.split("<|im_start|>assistant\n")[-1].split("<|im_end|>")[0]
+    return response
+
+demo = gr.ChatInterface(predict, type="messages")
+
+demo.launch()