Code

Supporting aimv2 encoders#2379

Code

Merged

rwightman merged 10 commits intomainhuggingface/pytorch-image-models:mainfrom aimv2huggingface/pytorch-image-models:aimv2Copy head branch name to clipboardDec 31, 2024

+518-49

Commits

Commits on Dec 27, 2024

• Supporting aimv2 encoders

  rwightman

  committed
  e752b5d

  Copy full SHA for e752b5d

Commits on Dec 29, 2024

• 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 'Lay…

  Show description for 04a484a
  rwightman

  committed
  04a484a

  Copy full SHA for 04a484a

• Switch aimv2 to used packed SwiGLU

  rwightman

  committed
  0d87cae

  Copy full SHA for 0d87cae

• Add bias arg to Vitamin GeGLU

  rwightman

  committed
  e0cacbf

  Copy full SHA for e0cacbf

• Use torch F.rms_norm when possible, select fast vs normal paths appropriately and test with torchscript

  rwightman

  committed
  5809c2f

  Copy full SHA for 5809c2f

Commits on Dec 30, 2024

• Add SimpleNorm to create_norm factory

  rwightman

  committed
  a4146b7

  Copy full SHA for a4146b7

• Add (almost) full set of aimv2 model instances. Switch back to unpacked SwiGLU. Verify correctness. Add DFN L/14 39B weight.

  rwightman

  committed
  1d6ebeb

  Copy full SHA for 1d6ebeb

• Update tests for aimv2 filtering

  rwightman

  committed
  cb294c8

  Copy full SHA for cb294c8

Commits on Dec 31, 2024

• test filter tweaks

  rwightman

  committed
  874037e

  Copy full SHA for 874037e

• Switch hf hub entries for new aimv2 / dfn weights to point to timm locations. Undo forced device for SDR linspace, part of another change.

  rwightman

  committed
  eb84e4b

  Copy full SHA for eb84e4b