Update app.py

app.py

@@ -1,64 +1,97 @@
+import torch
+from transformers import AutoProcessor, PaliGemmaForConditionalGeneration
+from PIL import Image
 import gradio as gr
-from huggingface_hub import InferenceClient
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+# -----------------------------------------------------------------------------
+# 1) GPU inference function
+# -----------------------------------------------------------------------------
+def run_inference_on_gpu(
+    model_id: str,
+    image: Image.Image,
+    prompt: str = "caption",
+    max_new_tokens: int = 100,
+    use_auth_token: bool = True
+) -> str:
+    # ensure CUDA is available
+    assert torch.cuda.is_available(), "CUDA not available—check your PyTorch installation!"
+    device = torch.device("cuda")
+    dtype = torch.float16
 
+    # load tokenizer + model onto GPU
+    processor = AutoProcessor.from_pretrained(model_id, use_auth_token=use_auth_token)
+    model = PaliGemmaForConditionalGeneration.from_pretrained(
+        model_id,
+        torch_dtype=dtype,
+        device_map=None,
+        use_auth_token=use_auth_token
+    ).to(device).eval()
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+    # build multimodal prompt
+    image_tokens = "<image>"
+    multimodal_prompt = f"{image_tokens} {prompt}"
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+    # prepare inputs
+    inputs = processor(
+        text=multimodal_prompt,
+        images=[image],
+        padding="longest",
+        return_tensors="pt",
+        do_convert_rgb=True,
+    )
+    inputs = {k: v.to(device) for k, v in inputs.items()}
 
-    messages.append({"role": "user", "content": message})
+    # generate
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=max_new_tokens,
+            num_beams=3,
+            do_sample=False,
+        )
 
-    response = ""
+    # decode
+    return processor.decode(outputs[0].cpu(), skip_special_tokens=True)
 
-    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
+# -----------------------------------------------------------------------------
+# 2) Gradio UI
+# -----------------------------------------------------------------------------
+MODEL_ID = "mychen76/paligemma-3b-mix-448-med_30k-ct-brain"
 
+def caption_fn(image, prompt, max_tokens):
+    """
+    Gradio callback: takes a PIL image, a text prompt, and
+    max tokens → returns the generated caption.
+    """
+    return run_inference_on_gpu(
+        model_id=MODEL_ID,
+        image=image,
+        prompt=prompt,
+        max_new_tokens=max_tokens,
+    )
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", 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"),
+demo = gr.Interface(
+    fn=caption_fn,
+    inputs=[
+        gr.Image(type="pil", label="Upload CT Scan"),
+        gr.Textbox(
+            value="What do you see in this CT scan?",
+            label="Prompt"
+        ),
         gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
+            minimum=10, maximum=300, step=10, value=100,
+            label="Max New Tokens"
         ),
     ],
+    outputs=gr.Textbox(label="Model Caption"),
+    title="PaliGemma CT-Scan Captioning",
+    description=(
+        "Upload a brain CT scan (or any image), write a short prompt, "
+        "and let the PaliGemma model describe what it sees."
+    ),
+    allow_flagging="never",
 )
 
-
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(share=False)  # set share=True if you need a public link