Create app.py

app.py

@@ -0,0 +1,27 @@
+import spaces
+import torch
+from transformers import AutoModel, AutoProcessor
+
+# Load di module level — ZeroGPU pakai CUDA emulation di luar @spaces.GPU
+processor = AutoProcessor.from_pretrained("TaichuAI/ZDTaichu5.0-9B", trust_remote_code=True)
+model = AutoModel.from_pretrained(
+    "TaichuAI/ZDTaichu5.0-9B",
+    torch_dtype=torch.bfloat16,
+    device_map="cuda",  # bukan "auto", ZeroGPU butuh explicit cuda
+    trust_remote_code=True,
+    attn_implementation="sdpa",
+).eval()
+
+@spaces.GPU(duration=60)  # default 60s, bisa naik ke 120s
+def generate(image, text):
+    messages = [{"role": "user", "content": [
+        {"type": "image", "image": image},
+        {"type": "text", "text": text},
+    ]}]
+    inputs = processor.from_messages(messages, return_tensors="pt").to("cuda")
+    with torch.inference_mode():
+        out = model.generate(**inputs, max_new_tokens=512, do_sample=False)
+    return processor.decode(out[0], skip_special_tokens=True)
+
+import gradio as gr
+gr.Interface(fn=generate, inputs=[gr.Image(), gr.Text()], outputs="text").launch()