Create app.py

app.py

@@ -0,0 +1,37 @@
+import gradio as gr
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+TOKENIZER_REPO = "MediaTek-Research/Breeze-7B-Instruct-v1_0"
+tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_REPO)
+model = AutoModelForCausalLM.from_pretrained(
+    TOKENIZER_REPO,
+    device_map="auto",
+    torch_dtype=torch.bfloat16,
+)
+
+
+
+def generate(text):
+    chat_data = []
+    text = text.strip()
+    if text:
+       chat_data.append({"role": "system", "content": text})
+    outputs = model.generate(tokenizer.apply_chat_template(chat, return_tensors="pt"),
+                         max_new_tokens=128,
+                         top_p=0.01,
+                         top_k=85,
+                         repetition_penalty=1.1,
+                         temperature=0.01)
+                         
+   print(tokenizer.decode(outputs[0]))
+   return tokenizer.decode(outputs[0])
+
+gradio_app = gr.Interface(
+    generate,
+    inputs=gr.Text(),
+    outputs=gr.Text(),
+    title="test",
+)
+
+if __name__ == "__main__":
+    gradio_app.launch()