Ernie4.5-VL-Video2Coder

Sleeping

zhijun.li commited on 22 days ago

Commit

850ce03

1 Parent(s): 432f4e8

Adjusting annotations

Files changed (1) hide show

app.py CHANGED Viewed

@@ -166,7 +166,7 @@ with gr.Blocks(title="Ernie 4.5 Video2Code", theme=gr.themes.Soft()) as demo:
     # 修复2：将 open 设置为 True，默认展开
     with gr.Accordion("📚 Technical Capabilities of ERNIE 4.5-VL", open=True):
         gr.Markdown("""
-        This application is powered by **Baidu ERNIE 4.5 **, a state-of-the-art foundation model with specific enhancements for video understanding:
         *   **👁️ Multimodal Heterogeneous MoE**: Uses dedicated vision experts to process images and video frames without interfering with text generation capabilities.
         *   **⏳ 3D-RoPE Temporal Modeling**: Incorporates 3D Rotary Position Embeddings to independently encode temporal, width, and height information.

     # 修复2：将 open 设置为 True，默认展开
     with gr.Accordion("📚 Technical Capabilities of ERNIE 4.5-VL", open=True):
         gr.Markdown("""
+        This application is powered by **Baidu ERNIE 4.5**, a state-of-the-art foundation model with specific enhancements for video understanding:
         *   **👁️ Multimodal Heterogeneous MoE**: Uses dedicated vision experts to process images and video frames without interfering with text generation capabilities.
         *   **⏳ 3D-RoPE Temporal Modeling**: Incorporates 3D Rotary Position Embeddings to independently encode temporal, width, and height information.