2762 Commits

Author SHA1 Message Date
Ross Wightman
2ad75e8023 Fix issue w/ MAP attention mask and no patch_valid 2025-05-24 12:37:39 -07:00
Ross Wightman
d7d3538335 Add so400m model size for test, few tweaks. 2025-05-23 18:29:30 -07:00
Ross Wightman
7bfe606d9f Merge remote-tracking branch 'origin/main' into naflex 2025-05-23 09:41:49 -07:00
Ross Wightman
e7925ea6d6 Add explicit disable of fc_norm to pe giant spatial variant 2025-05-21 07:31:36 -07:00
Ross Wightman
f001b15ed3 NaFlex random erasing performance improvements, python loops were slow. Remove subregion mode, not going to be worth it. 2025-05-20 17:03:46 -07:00
Ross Wightman
7624389fc9 Mixup cleanup, add prob support and train script integration. Add working loader based patch compatible RandomErasing for NaFlex mode. 2025-05-20 14:38:50 -07:00
Ross Wightman
edc37be1c3 Remove default 1000 classes from onnx export to use model default 2025-05-15 14:14:53 -07:00
Ross Wightman
f14f6507ab
Merge pull request #2487 from huggingface/eva_pe_integration
Add EVA ViT based PE (Perceptual Encoder) impl
2025-05-15 12:56:03 -07:00
Ross Wightman
cabd26d018 Fix missing device arg in train.py, fix #2488 2025-05-14 13:51:43 -07:00
Ross Wightman
6b302f27a3
Merge pull request #2480 from brianhou0208/tnt
Update TNT-(S/B) model weights and add feature extraction support
2025-05-14 12:27:21 -07:00
Ross Wightman
16d0b26e19 Fix torchscript issue with legacy tnt 2025-05-14 09:33:41 -07:00
Ross Wightman
74ad32a67e Updated tnt model weights on hub, add back legacy model in case bwd compat 2025-05-14 08:40:53 -07:00
Ryan
69b1fbcdc1 Merge branch 'main' into tnt 2025-05-11 22:45:48 +08:00
Ross Wightman
88b7ef6035 Disable dynamic_img_size default on PE models for now 2025-05-10 15:00:29 -07:00
Ross Wightman
8fcbceb609 Add a WIP NaFlex compatible mixup/cutmix for testing 2025-05-10 14:59:37 -07:00
Ross Wightman
fa7287faee Add EVA ViT based PE (Perceptual Encoder) impl 2025-05-10 13:13:12 -07:00
Ryan Ramos
81900a6bae remove download argument from torch_kwargs 2025-05-09 07:34:32 -07:00
atharva-pathak
59e59ff040 Update README.md
The previous link to Chris Hughes blog post was wrong, so updated the correct link
2025-05-08 08:43:18 -07:00
Ryan
d1140c1a0f fix pit & add to test 2025-05-08 07:17:19 -07:00
Ryan
2e9b2a76fb update some model 2025-05-08 07:17:19 -07:00
Ryan
99c25fa5c0 fix nest type error 2025-05-08 07:17:19 -07:00
Ryan
72b2a09106 fix final norm only apply at last indice 2025-05-08 07:17:19 -07:00
Ryan
e0ae4db8fc update FEAT_INTER_FILTERS list 2025-05-08 07:17:19 -07:00
Ryan
98d9cde8b4 support efficientformer_v2 2025-05-08 07:17:19 -07:00
Ryan
12def0d118 support efficientvit, edgenext, davit 2025-05-08 07:17:19 -07:00
Ryan
9aedecbb5f fix norm at last feat_idx 2025-05-08 07:17:19 -07:00
Ryan
e16d385592 support more models
inception_next, hgnet, gcvit, focalnet, inception_v4
2025-05-08 07:17:19 -07:00
Ryan
6b6beffa6b support mambaout, metaformer, nest, nextvit, pvt_v2 2025-05-08 07:17:19 -07:00
Ryan
2d3155908c fix tresnet and rdnet 2025-05-08 07:17:19 -07:00
Ryan
f8be741f0f support rexnet, resnetv2, repvit and repghostnet 2025-05-08 07:17:19 -07:00
Ryan
5e8cc616d4 support tiny_vit 2025-05-08 07:17:19 -07:00
Ryan
8befebd93c support vovnet 2025-05-08 07:17:19 -07:00
Ryan
411b892dbc support tresnet 2025-05-08 07:17:19 -07:00
Ross Wightman
fdcf7cf5c3 Check forward_intermediates features against forward_features output 2025-05-06 13:48:51 -07:00
Ryan
37bbac1256 Fix checkpoint_filter_fn 2025-05-03 15:41:18 +08:00
Ryan
fc0b6ad183 Fix default_cfgs 2025-05-03 08:40:07 +08:00
Ryan
848b8c3e57 Support features_only 2025-05-02 20:59:05 +08:00
Ryan
b37f0f7a76 Update tnt.py 2025-05-02 20:34:31 +08:00
Ross Wightman
e2073e32d0 Move NaFlexCollate with dataset, remove stand alone collate_fn and remove redundancy 2025-04-29 10:44:46 -07:00
Ross Wightman
39eb56f875 Starting to test distributed train, fix issue with batch_size reduce 2025-04-28 16:48:06 -07:00
Ross Wightman
ee27b73da4 Further pos embed tweaks, rejig model defs for testing 2025-04-28 09:15:11 -07:00
Ross Wightman
3dc90ed7a7 Add naflex loader support to validate.py, fix bug in naflex pos embed add, classic vit weight loading for naflex model 2025-04-25 16:00:54 -07:00
Yutong Xiang
c8c4f256b8
add-lightly-train-to-training-frameworks-readme (#2474) 2025-04-22 08:50:11 -07:00
Ross Wightman
c527c37969 Optimizations for pos embed resize, merge different mask helper fns 2025-04-21 14:05:18 -07:00
Xihua Dong
0c136f7dab fix img_size type 2025-04-18 21:05:05 -07:00
Ross Wightman
3ff3899026 Add local-dir: schema support for model loading (config + weights) from folder 2025-04-17 12:32:19 -07:00
Ross Wightman
ceca5efdec Remove torch_out from onnx export, no point without the export_ fn 2025-04-15 14:01:17 -07:00
Ross Wightman
0cae8a4cd8 Fix #2472, torch.onnx.export_ (with return output) finally removed :( 2025-04-15 14:01:17 -07:00
Ross Wightman
ea728f67fa Improve several typing issues for flex vit, can (almost) work with jit if we bash h,w key into an int or str 2025-04-14 11:01:56 -07:00
Ross Wightman
681be882e8 Fix arg merging of sknet, old seresnet. Fix #2470 2025-04-14 10:32:26 -07:00