Disable dynamic_img_size default on PE models for now

Add a WIP NaFlex compatible mixup/cutmix for testing

Add EVA ViT based PE (Perceptual Encoder) impl

remove download argument from torch_kwargs

Update README.md

fix pit & add to test

Check forward_intermediates features against forward_features output

Check forward_intermediates features against forward_features output

Move NaFlexCollate with dataset, remove stand alone collate_fn and re…

Starting to test distributed train, fix issue with batch_size reduce

Further pos embed tweaks, rejig model defs for testing

Add naflex loader support to validate.py, fix bug in naflex pos embed…

add-lightly-train-to-training-frameworks-readme (#2474)

Optimizations for pos embed resize, merge different mask helper fns

fix img_size type

Add local-dir: schema support for model loading (config + weights) fr…

Add local-dir: schema support for model loading (config + weights) fr…

Remove torch_out from onnx export, no point without the export_ fn

Remove torch_out from onnx export, no point without the export_ fn

Remove torch_out from onnx export, no point without the export_ fn

Fix #2472, torch.onnx.export_ (with return output) finally removed :(

Improve several typing issues for flex vit, can (almost) work with ji…

Fix arg merging of sknet, old seresnet. Fix #2470

Fix arg merging of sknet, old seresnet. Fix #2470

A much faster resample_patch_embed, can be used at train/validation time

Exclude embeds module and mask attn functions from tracing

Add loss scale arg, initial distributed loss scale. Maybe fix FX for …

Fix ParallelThingsBlock w/ attn_mask

Exclude naflex models from jit tests

Type fixes, remove old comments