frichter comments

Repositories
Issues
Comments

Results 2 comments of


                                            frichter

Anyone tested quantization with success? my results are bad

See my answer here: https://github.com/facebookresearch/dinov3/issues/288#issuecomment-3672045590

Patch embeddings are all the same with float16 on mps

Hello! I noticed a similar problem when quantizing our model (which uses DinoV3 as a backbone) to FP16. I traced the problem back to the first layer of the model...