1756 Commits

Author	SHA1	Message	Date
Ross Wightman	5d535d7a2d	Version 1.0.14, update README & changelog	2025-01-19 13:53:09 -08:00
Ross Wightman	aa333079da	Tweak so150m2 def	2025-01-19 13:40:53 -08:00
Josua Rieder	8d81fdf3d9	Fix typos	2025-01-19 13:39:40 -08:00
Ross Wightman	3677f67902	Add the 256x256 in1k ft of the so150m, add an alternate so150m def	2025-01-18 15:51:57 -08:00
Ross Wightman	2a84d68d02	Add some so150m vit w/ sbb recipe weights, and a ese_vovnet57b model with RA4 recipe	2025-01-18 15:51:57 -08:00
Ross Wightman	9265d54a3a	LeViT safetensors load is broken by conversion code that wasn't deactivated	2025-01-16 11:37:00 -08:00
Ross Wightman	21e75a9d25	Update version.py ... Back to dev version	2025-01-16 11:23:17 -08:00
Adam J. Stewart	6d21eb0d37	VGG ConvMlp: fix layer defaults/types	2025-01-15 12:11:56 +01:00
Adam J. Stewart	f5c4d5cbb7	Add missing imports	2025-01-11 15:13:16 +01:00
Adam J. Stewart	19aaea3c8f	Fix nn.Module type hints	2025-01-11 15:09:21 +01:00
Ross Wightman	47811bc05a	Update README, bump version to 1.0.13 non-dev	2025-01-09 09:33:59 -08:00
Ross Wightman	deb9895600	Update checkpoint save to fix old hard-link + fuse issue I ran into again... fix #340	2025-01-08 15:36:58 -08:00
Ross Wightman	92f610c982	Add half-precision (bfloat16, float16) support to train & validate scripts. Should push dtype handling into model factory / pretrained load at some point...	2025-01-07 10:25:14 -08:00
Ross Wightman	155f6e7fea	Update README, few minor fixups.	2025-01-06 13:09:15 -08:00
Ross Wightman	2b251fb291	Wrap torch checkpoint() fn to default use_reentrant flag to False and allow env var override	2025-01-06 11:28:39 -08:00
Ross Wightman	131518c15c	Add comments to MLP layers re expected layouts	2025-01-02 09:41:35 -08:00
Louis Lac	2d5277e858	Merge branch 'main' into fix-mqa-v2	2025-01-02 00:11:22 +01:00
Louis Lac	2d734d9058	Fixed unfused attn2d scale	2025-01-01 12:34:07 -08:00
Louis Lac	6171e756d3	Fix MQA V2 scale and out shape	2025-01-01 15:37:28 +01:00
Ross Wightman	e846b2cf28	Add 384x384 in12k pretrain and finetune for convnext_nano	2024-12-31 13:16:43 -08:00
Ross Wightman	b0068ba5d0	Switch hf hub entries for new aimv2 / dfn weights to point to timm locations. Undo forced device for SDR linspace, part of another change.	2024-12-30 19:24:21 -08:00
Ross Wightman	1bf84b35c3	Update tests for aimv2 filtering	2024-12-30 19:24:21 -08:00
Ross Wightman	b33418713a	Add (almost) full set of aimv2 model instances. Switch back to unpacked SwiGLU. Verify correctness. Add DFN L/14 39B weight.	2024-12-30 19:24:21 -08:00
Ross Wightman	de35fd87f5	Add SimpleNorm to create_norm factory	2024-12-30 19:24:21 -08:00
Ross Wightman	d5375ca769	Use torch F.rms_norm when possible, select fast vs normal paths appropriately and test with torchscript	2024-12-30 19:24:21 -08:00
Ross Wightman	5f12a25114	Add bias arg to Vitamin GeGLU	2024-12-30 19:24:21 -08:00
Ross Wightman	5804d92e4b	Switch aimv2 to used packed SwiGLU	2024-12-30 19:24:21 -08:00
Ross Wightman	15406a939e	Fixing RmsNorm to fix #2380 and noticed with aimv2 when comparing outputs. Still some work to do, need to look at AMP / fast mode behaviour, dispatch to torch when possible. Add SimpleNorm for 'LayerNorm w/o centering and bias'	2024-12-30 19:24:21 -08:00
Ross Wightman	a648a04834	Supporting aimv2 encoders	2024-12-30 19:24:21 -08:00
Ross Wightman	790decc89b	Add more pali(2) weights. Switch rest of models adapting open_clip weights to their own weight instances.	2024-12-27 14:00:41 -08:00
Ross Wightman	01cf0f72af	Add support for tag, license customization through push_to_hub	2024-12-27 14:00:41 -08:00
Ross Wightman	b12ecbd614	Move siglip timm weights to own repos	2024-12-27 14:00:41 -08:00
Ross Wightman	6fb7aaf37d	Switching to timm specific weight instances for open_clip image encoders to facilitate hf-hub: use in timm and new transformers TimmWrapper	2024-12-27 14:00:41 -08:00
Ross Wightman	364c567dd2	Merge pull request #2357 from huggingface/more_opt_stuff ... Add caution to Adan. Add decouple decay option to LAMB.	2024-12-27 12:54:02 -08:00
Ryan	ab0a70dfff	fix feature_info.reduction	2024-12-18 21:12:40 +08:00
Ross Wightman	7573096eb8	Make sure trust_remote code only passed to HF datasets. Improve some docstrings.	2024-12-06 11:40:04 -08:00
Ross Wightman	9eee47de52	Back to dev version	2024-12-06 10:44:41 -08:00
Álvaro Justen (@turicas)	9383f2880d	Add cache_dir example	2024-12-06 10:39:13 -08:00
Ross Wightman	d1e9a8622a	Rename inception_next_atto pretrained str	2024-12-06 10:36:47 -08:00
Weihao Yu	0576175d85	Add inception_next_atto	2024-12-06 10:36:47 -08:00
Ross Wightman	7ab2b938e5	More tweaks to docstrings for hub/builder	2024-12-06 10:25:06 -08:00
Ross Wightman	dc1bb05e8e	Punch cache_dir through model factory / builder / pretrain helpers. Improve some annotations in related code.	2024-12-06 10:25:06 -08:00
Ross Wightman	afdf11d9ae	Add caution to Adan. Add decouple decay option to LAMB.	2024-12-05 13:50:30 -08:00
Ross Wightman	553ded5c6b	Version 1.0.12	2024-12-03 10:34:52 -08:00
Ross Wightman	464885e135	See if we can avoid some model / layer pickle issues with the aa attr in ConvNormAct	2024-12-03 08:02:55 -08:00
Ross Wightman	5fe5f9d488	Add a different mnv4 conv-small weight	2024-12-02 16:14:37 -08:00
Ross Wightman	303f7691a1	Add cautious mars, improve test reliability by skipping grad diff for first step	2024-12-02 11:29:02 -08:00
Ross Wightman	82e8677690	Make LaProp weight decay match typical PyTorch 'decoupled' behaviour where it's scaled by LR	2024-11-29 16:44:43 -08:00
Ross Wightman	886eb77938	Update README, missed small discrep in adafactor min dim update	2024-11-29 10:57:47 -08:00
Ross Wightman	e3e434bbc4	To be technically correct, need to check the in-place _ ver of op	2024-11-28 15:11:58 -08:00