Florence 2 Fine-Tuning: How to Train a Vision Language Model?

[@MervinPraison]

* Not Flowerence , it's Florence

[@unclecode]

As always, brief and to the point. A question: Have you tested after these 1000 epochs and frozen the image encoder? Curious to know how user-friendly the fine-tuning process is.

[@miket64]

Great job, your videos are very inspiring, thanks!

[@user-uk9ls]

Excellent tutorial

[@Ravensd-x8t]

How to use the fine tuned model, it shows "only DaViT is supported for now".when i try to use i got this error 😢😢

[@sudabadri7051]

What i was waiting for

[@tonywhite4476]

I see you've been reduced to promo videos for big tech. GL with that.

[@Nick_With_A_Stick]

Do you need the gradients of the inputs during training?

[@swetanshuprasad9103]

where is the link?

[@redbaron3555]

Habe you tried Grokking it with a lot of Epochs?

[@avivgraupen6707]

The code runs only on one GPU. Do u have a code version for multi GPU ?

[@free_thinker4958]

Can we run florence 2 in cpu??

[@redbaron3555]

You can run moondream on an M1 or M2

[@redpen9515]

How can you fine-tune a model without testing it? Additionally, the video thumbnail doesn't match the content you taught.

[@illiyask-e8d]

why givng error when load this model from hugginface. " 2536 self.vision_tower = DaViT.from_config(config=config.vision_config) 2537 # remove unused layers AssertionError: only DaViT is supported for now "