Commit Graph

  • 9d294cd046 Merge pull request #2510 from emmanuel-ferdman/main Ross Wightman 2025-06-09 21:09:56 -07:00
  • 3f543f1137 Throw exception on non-directory path for pretrained weights Emmanuel Ferdman 2025-06-09 16:20:18 -07:00
  • 6a621b5b1c Pass block_fn and mlp_layer through from NaFlexVit cfg, fixes a few models Ross Wightman 2025-06-07 13:58:37 -07:00
  • f9b3d7e00c Merge pull request #2507 from huggingface/more_naflex Ross Wightman 2025-06-05 12:55:18 -07:00
  • 6c7ce457bf Update README.md with some more NaFlexVit details more_naflex Ross Wightman 2025-06-05 12:54:40 -07:00
  • 4d9c87bbb2 Fix polarity on prefix remove Ross Wightman 2025-06-05 12:17:59 -07:00
  • cb3e5b72b8 Forgot to compact attention pool branches after verifying Ross Wightman 2025-06-05 12:08:24 -07:00
  • 1299488303 dev version Ross Wightman 2025-06-05 10:38:50 -07:00
  • a5e551b0f4 Merge pull request #2466 from huggingface/naflex Ross Wightman 2025-06-05 09:12:37 -07:00
  • a0b5bcc558 Fix another low use path where only numpy arrays are supported naflex Ross Wightman 2025-06-05 08:32:58 -07:00
  • 99a09eb843 Update old FastCollateMixup to accept torch tensor inputs instead of only numpy arrays Ross Wightman 2025-06-05 08:25:27 -07:00
  • 4ff865caec A bit of docstring and comment consistency cleanup, remove some debug code Ross Wightman 2025-06-04 21:11:13 -07:00
  • dac2ec686f Add missing patch embed interpolator Ross Wightman 2025-06-04 18:24:29 -07:00
  • 0d439427f2 Add variable patch size to naflex training, improve patch size arg handling from train.py onwards. Add docstrings and type annotations (thanks Claude). Ross Wightman 2025-06-04 17:03:46 -07:00
  • d78cbf474f Rename dataset wrapper to NaFlexMapDatasetWrapper Ross Wightman 2025-06-04 08:17:53 -07:00
  • dd3b96c755 Fix features intermediates for NCHW inputs, patch variable size inputs need more code Ross Wightman 2025-06-03 21:19:01 -07:00
  • b3ca8fd195 Add naflex vit exceptions to tests Ross Wightman 2025-06-03 17:16:20 -07:00
  • 2bf71f54e9 Merge remote-tracking branch 'origin/main' into naflex Ross Wightman 2025-06-03 16:28:30 -07:00
  • fe2867c710 Significant naflex refactor. Rename classes, models. Support flag for creating classic vits as naflex. Cleanup, improvements. Ross Wightman 2025-06-03 15:48:41 -07:00
  • a22366e3ce Merge pull request #2503 from huggingface/beit3_remap_clean Ross Wightman 2025-05-30 16:40:28 -07:00
  • 72858c193c Add siglip2 compatible naflex encoders. Add support to factorized pos embeds and 'aspect preserving mode' to Flex Embeds. Some more docstrings and typing. Ross Wightman 2025-05-30 16:15:37 -07:00
  • 211cf90721 Imports getting unwieldy in vision_transformer.py beit3_remap_clean Ross Wightman 2025-05-30 15:11:51 -07:00
  • 1be7999993 Upload beit3 weights to hub, add pretrain weights Ross Wightman 2025-05-30 14:36:41 -07:00
  • 3a3d98bc38 Fix parallel blocks missing scale args and vitamin MLP Ross Wightman 2025-05-29 11:34:19 -07:00
  • 2ca94a6ce4 Compact _covert_beit3 fn Ross Wightman 2025-05-29 10:52:39 -07:00
  • 38c5f3bc63 A cleaned up beit3 remap onto vision_transformer.py vit Ross Wightman 2025-05-29 10:33:09 -07:00
  • 55e52c45ef Initial run through remapping beit3 -> vision_transformer.py beit3_remap Ross Wightman 2025-05-29 09:50:17 -07:00
  • b0b28e29aa fix metaformer and nextvit Ryan 2025-05-29 05:17:16 +08:00
  • a80348a8c8 support starnet and ghostnet Ryan 2025-05-29 05:13:50 +08:00
  • 57f85542da support gradient checkpoint in forward_intermediates Ryan 2025-05-29 05:00:38 +08:00
  • a0a30a6b48 Update README.md Ross Wightman 2025-05-28 09:42:09 -07:00
  • 3114a9f595 Merge pull request #2499 from brianhou0208/fast_model Ross Wightman 2025-05-26 18:49:51 -07:00
  • 4209788233 Updated faster models w/ hub weight locations, commented out some checkpoint filter fns and minor renames Ross Wightman 2025-05-26 16:00:29 -07:00
  • b7ced7c40c torch.fx.wrap not working with older pytorch, trying register_notrace instead Ross Wightman 2025-05-25 14:13:36 -07:00
  • 842a786626 A few more maybe_add_mask situations Ross Wightman 2025-05-25 08:51:56 -07:00
  • dd2c1418d0 Fix tracing of attention module with attn_mask support Ross Wightman 2025-05-24 21:13:01 -07:00
  • 162f49295e Move naflex global pool into one fn that can be marked notrace Ross Wightman 2025-05-24 14:06:12 -07:00
  • 2ad75e8023 Fix issue w/ MAP attention mask and no patch_valid Ross Wightman 2025-05-24 12:37:39 -07:00
  • d7d3538335 Add so400m model size for test, few tweaks. Ross Wightman 2025-05-23 18:29:30 -07:00
  • 7bfe606d9f Merge remote-tracking branch 'origin/main' into naflex Ross Wightman 2025-05-23 09:41:49 -07:00
  • 75823ab90e Merge branch 'main' into fast_model Ryan 2025-05-22 02:37:00 +08:00
  • e7925ea6d6 Add explicit disable of fc_norm to pe giant spatial variant Ross Wightman 2025-05-21 07:31:36 -07:00
  • f001b15ed3 NaFlex random erasing performance improvements, python loops were slow. Remove subregion mode, not going to be worth it. Ross Wightman 2025-05-20 17:03:46 -07:00
  • 7624389fc9 Mixup cleanup, add prob support and train script integration. Add working loader based patch compatible RandomErasing for NaFlex mode. Ross Wightman 2025-05-20 14:38:03 -07:00
  • 9790fea406 Merge branch 'main' into beit3 Ryan 2025-05-17 22:31:07 +08:00
  • bea1137583 Merge branch 'main' into fast_model Ryan 2025-05-17 22:25:34 +08:00
  • 236b00cf2d add ghostnetv3 Ryan 2025-05-17 17:57:54 +08:00
  • edc37be1c3 Remove default 1000 classes from onnx export to use model default Ross Wightman 2025-05-15 14:14:53 -07:00
  • f14f6507ab Merge pull request #2487 from huggingface/eva_pe_integration Ross Wightman 2025-05-15 12:56:03 -07:00
  • cabd26d018 Fix missing device arg in train.py, fix #2488 Ross Wightman 2025-05-14 13:51:43 -07:00
  • 6b302f27a3 Merge pull request #2480 from brianhou0208/tnt Ross Wightman 2025-05-14 12:27:21 -07:00
  • 1922ca5f1b fix fasternet Ryan 2025-05-15 02:28:17 +08:00
  • 16d0b26e19 Fix torchscript issue with legacy tnt Ross Wightman 2025-05-14 09:33:41 -07:00
  • 74ad32a67e Updated tnt model weights on hub, add back legacy model in case bwd compat Ross Wightman 2025-05-14 08:40:53 -07:00
  • cb98094792 Merge branch 'main' into fast_model Ryan 2025-05-14 08:30:33 +08:00
  • 89d2952375 update group_matcher Ryan 2025-05-14 08:28:10 +08:00
  • b5a814e4c1 add giant model param Ryan 2025-05-12 00:24:15 +08:00
  • afe4375e77 update BEiT3 Ryan 2025-05-12 00:13:52 +08:00
  • 008514934c Merge branch 'main' into beit3 Ryan 2025-05-11 22:56:10 +08:00
  • 69b1fbcdc1 Merge branch 'main' into tnt Ryan 2025-05-11 22:45:48 +08:00
  • 88b7ef6035 Disable dynamic_img_size default on PE models for now eva_pe_integration Ross Wightman 2025-05-10 15:00:29 -07:00
  • 8fcbceb609 Add a WIP NaFlex compatible mixup/cutmix for testing Ross Wightman 2025-05-10 14:59:37 -07:00
  • fa7287faee Add EVA ViT based PE (Perceptual Encoder) impl Ross Wightman 2025-05-10 13:13:12 -07:00
  • 81900a6bae remove download argument from torch_kwargs Ryan Ramos 2025-05-09 15:11:22 +09:00
  • 59e59ff040 Update README.md atharva-pathak 2025-05-08 10:41:15 +05:30
  • d1140c1a0f fix pit & add to test Ryan 2025-05-08 02:02:54 +08:00
  • 2e9b2a76fb update some model Ryan 2025-05-08 00:57:18 +08:00
  • 99c25fa5c0 fix nest type error Ryan 2025-05-06 03:43:49 +08:00
  • 72b2a09106 fix final norm only apply at last indice Ryan 2025-05-06 00:56:36 +08:00
  • e0ae4db8fc update FEAT_INTER_FILTERS list Ryan 2025-05-06 00:25:53 +08:00
  • 98d9cde8b4 support efficientformer_v2 Ryan 2025-05-06 00:25:21 +08:00
  • 12def0d118 support efficientvit, edgenext, davit Ryan 2025-05-06 00:24:57 +08:00
  • 9aedecbb5f fix norm at last feat_idx Ryan 2025-05-05 23:15:39 +08:00
  • e16d385592 support more models Ryan 2025-05-05 23:11:59 +08:00
  • 6b6beffa6b support mambaout, metaformer, nest, nextvit, pvt_v2 Ryan 2025-05-05 05:30:57 +08:00
  • 2d3155908c fix tresnet and rdnet Ryan 2025-05-05 04:22:11 +08:00
  • f8be741f0f support rexnet, resnetv2, repvit and repghostnet Ryan 2025-05-05 04:21:12 +08:00
  • 5e8cc616d4 support tiny_vit Ryan 2025-05-05 01:35:02 +08:00
  • 8befebd93c support vovnet Ryan 2025-05-04 23:49:58 +08:00
  • 411b892dbc support tresnet Ryan 2025-05-04 23:49:43 +08:00
  • fdcf7cf5c3 Check forward_intermediates features against forward_features output Ross Wightman 2025-05-06 12:56:58 -07:00
  • 907a32e699 Check forward_intermediates features against forward_features output check_intermediate_feat Ross Wightman 2025-05-06 12:56:58 -07:00
  • 7fc0692843 update fasternet Ryan 2025-05-06 03:10:19 +08:00
  • a3e66b14ea add fasternet Ryan 2025-05-06 02:59:33 +08:00
  • 7aeebf20e2 add BEIT3 Ryan 2025-05-06 01:29:55 +08:00
  • 91e6e1737e add test_models Ryan 2025-05-03 23:40:37 +08:00
  • 37bbac1256 Fix checkpoint_filter_fn Ryan 2025-05-03 15:41:18 +08:00
  • fc0b6ad183 Fix default_cfgs Ryan 2025-05-03 08:40:07 +08:00
  • 848b8c3e57 Support features_only Ryan 2025-05-02 20:59:05 +08:00
  • b37f0f7a76 Update tnt.py Ryan 2025-05-02 20:34:31 +08:00
  • e2073e32d0 Move NaFlexCollate with dataset, remove stand alone collate_fn and remove redundancy Ross Wightman 2025-04-29 10:44:46 -07:00
  • 39eb56f875 Starting to test distributed train, fix issue with batch_size reduce Ross Wightman 2025-04-28 16:48:06 -07:00
  • 77788f4f92 add shvit Ryan 2025-04-29 06:24:44 +08:00
  • b60a4fb329 add starnet Ryan 2025-04-29 05:38:42 +08:00
  • 081e6c2b96 update swiftformer Ryan 2025-04-29 05:19:53 +08:00
  • 7a87199bf8 add swiftformer Ryan 2025-04-29 00:58:10 +08:00
  • ee27b73da4 Further pos embed tweaks, rejig model defs for testing Ross Wightman 2025-04-28 09:15:11 -07:00
  • 3dc90ed7a7 Add naflex loader support to validate.py, fix bug in naflex pos embed add, classic vit weight loading for naflex model Ross Wightman 2025-04-25 16:00:54 -07:00
  • c8c4f256b8 add-lightly-train-to-training-frameworks-readme (#2474) Yutong Xiang 2025-04-22 17:50:11 +02:00
  • c527c37969 Optimizations for pos embed resize, merge different mask helper fns Ross Wightman 2025-04-21 14:05:18 -07:00