[MODEL REQUEST] Florence-2

[EwoutH]

Details of model being requested

• Model name: Florence-2
• Source repo link: https://huggingface.co/collections/microsoft/florence-6669f44df0d87d9c3bfb76de
• Research paper link: https://arxiv.org/abs/2311.06242
• Model use case: Florence-2 is an advanced vision foundation model designed for a wide range of vision and vision-language tasks. Utilizing a prompt-based approach, Florence-2 excels in image captioning, object detection, and segmentation. Its sequence-to-sequence architecture enables high performance in both zero-shot and fine-tuned settings. The model is trained on the extensive FLD-5B dataset, which includes 5.4 billion annotations across 126 million images, making it highly effective for multi-task learning.

Model Variants

Model	Model size	Model Description
Florence-2-base	0.23B	Pretrained model with FLD-5B
Florence-2-large	0.77B	Pretrained model with FLD-5B
Florence-2-base-ft	0.23B	Finetuned model on a collection of downstream tasks
Florence-2-large-ft	0.77B	Finetuned model on a collection of downstream tasks

Online demo: https://huggingface.co/spaces/gokaygokay/Florence-2

[mestrona-3]

Hi @EwoutH, thank you for the feature request and details! I've added it to our list of requested models.

If you haven't already, we'd invite you to join our AI Hub Slack Community as that is where we share new models, features and talk with model developers using AI Hub.

[EwoutH]

@mestrona-3 could you give any update on model additions currently in the pipeline? I would love Florence-2, Llama 3.1 and Phi-3 to be available on the Qualcomm AI Hub.

[mestrona-3]

Hi @EwoutH, we're working on many LLMs to be available via AI Hub. We'll share as soon as we can, via Slack channel