| library_name: pytorch | |
|  | |
| Florence-2 is a unified vision foundation model that leverages prompt-based learning to perform a wide range of vision and vision-language tasks using a single architecture and training framework. | |
| Original paper: [Advancing a Unified Representation for a Variety of Vision Tasks](https://arxiv.org/abs/2311.06242) | |
| # Florence-2-base | |
| This model uses the Florence-2 Base variant, which provides a balance between accuracy and computational efficiency while supporting multiple tasks through natural language prompts. It is well suited for applications such as image captioning, visual question answering, object detection, grounding, and general-purpose vision understanding. | |
| Model Configuration: | |
| - Reference implementation: [Florence-2](https://github.com/microsoft/Florence-2) | |
| - Original Weight: [Florence-2-base](https://huggingface.co/microsoft/Florence-2-base) | |
| - Resolution: 3x768x768 (3x384x384 on CV75) | |
| - Support Cooper version: | |
| - Cooper SDK: [2.5.4] | |
| - Cooper Foundry: [2.3] | |
| | Model | Device | compression | Model Link | | |
| | :---------------: | :------: | :-------------: | :------------------------------------------------------------------------------------------------: | | |
| | Florence-2-base | N1-655 | 8-bit weights | [Model_Link](https://huggingface.co/Ambarella/Florence2/blob/main/n1-655_florence2_0.23B.tar.gz) | | |
| | Florence-2-base | CV7 | 8-bit weights | [Model_Link](https://huggingface.co/Ambarella/Florence2/blob/main/cv7_florence2_0.23B.tar.gz) | | |
| | Florence-2-base | CV72 | 8-bit weights | [Model_Link](https://huggingface.co/Ambarella/Florence2/blob/main/cv72_florence2_0.23B.tar.gz) | | |
| | Florence-2-base | CV75 | 8-bit weights | [Model_Link](https://huggingface.co/Ambarella/Florence2/blob/main/cv75_florence2_0.23B.tar.gz) | | |