# Model checkpoints for [PMF](https://github.com/hushell/pmf_cvpr22)

NOTE: for DINO-small, peak VRAM is about 32GB; for DINO-base, peak VRAM is about 42GB.

Meta-testing with `dino_small_batch16` trained on full Meta-Dataset:
```
python -m torch.distributed.launch --nproc_per_node=8 --use_env test_meta_dataset.py --data-path ../../datasets/meta_dataset --dataset meta_dataset --arch dino_small_patch16 --deploy finetune --output outputs/md_full_dinosmall --resume md_full_128x128_dinosmall_fp16_lr5e-5/best.pth --dist-eval --ada_steps 100 --ada_lr 0.0001
```

Meta-testing with `dino_small_batch16` trained on ImageNet domain of Meta-Dataset:
```
python -m torch.distributed.launch --nproc_per_node=8 --use_env test_meta_dataset.py --data-path ../../datasets/meta_dataset --dataset meta_dataset --arch dino_small_patch16 --deploy finetune --output outputs/md_inet_dinosmall_6gpus --resume pmf_metadataset_dino/md_inet_128x128_dinosmall_fp16_lr5e-5/best.pth --dist-eval --ada_steps 100 --ada_lr 0.0001
```