Spaces:

HuggingFaceM4
/

idefics3

Running on Zero

merve HF staff commited on Aug 6

Commit

959aec5

•

1 Parent(s): b926faa

Update app.py

Files changed (1) hide show

app.py CHANGED Viewed

@@ -81,6 +81,7 @@ def model_inference(
 with gr.Blocks(fill_height=True) as demo:
     gr.Markdown("## IDEFICS3-Llama 🐶")
     gr.Markdown("Play with [HuggingFaceM4/Idefics3-8B-Llama3](https://huggingface.co/HuggingFaceM4/Idefics3-8B-Llama3) in this demo. To get started, upload an image and text or try one of the examples.")
     with gr.Column():
         image_input = gr.Image(label="Upload your Image", type="pil")
         query_input = gr.Textbox(label="Prompt")

 with gr.Blocks(fill_height=True) as demo:
     gr.Markdown("## IDEFICS3-Llama 🐶")
     gr.Markdown("Play with [HuggingFaceM4/Idefics3-8B-Llama3](https://huggingface.co/HuggingFaceM4/Idefics3-8B-Llama3) in this demo. To get started, upload an image and text or try one of the examples.")
+    gr.Markdown("**Disclaimer:** Idefics3 does not include an RLHF alignment stage, so it may not consistently follow prompts or handle complex tasks. However, this doesn't mean it is incapable of doing so. Adding a prefix to the assistant's response, such as Let's think step for a reasoning question or <html> for HTML code generation, can significantly improve the output in practice. You could also play with the parameters such as the temperature in non-greedy mode.")
     with gr.Column():
         image_input = gr.Image(label="Upload your Image", type="pil")
         query_input = gr.Textbox(label="Prompt")