guarin

Thanks for the issue! Was able to reproduce this on the main branch:  Ran it with: ``` python main.py --train-dir /datasets/imagenet1k/train --val-dir /datasets/imagenet1k/val/ --epochs 0 --ckpt-path ../../../epoch\=99-step\=500400.ckpt --skip-linear-eval --skip-finetune-eval...

Looks like this might be related to the changes we introduced in #1800 I tried with commit `b6955fd40b9b8e2f11cbd6d291820281ed47ba3a` (v1.5.18) and got the expected results: So the checkpoint weights are good...

Hi @EricLiclair, the idea is that the `main.py` script in the benchmarks should aggregate all evaluation metrics and print them as a table in the end. The relevant function calls...

> Also are you on the hacktober's discord? (If it's comfortable with you to disclose that? If I had some doubts, I thought I could contact you their itself) Yes...

Completed in #1680

The default is to use a single linear layer that projects the student features to the same dimension as the teacher (`nn.Linear(512, 4096)`)

> Hi @yutong-xiang-97 I noticed that you are distilling from the CLS token's embedding outputted by DINOv3. I was wondering if we can choose to distill using the entire feature...

You are right that `forward` returns only CLS tokens when the model is in `eval` mode. But we call `get_intermediate_layers` which always returns feature maps. > Also, I am curious,...

Yes it is much stronger than ImageNet pretrained

Yes you can definitely do this. We actually use features from multiple DINO blocks for distillation.