Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 3, 2021

• Add ResNet101 and ResNet152 weights from higher aug RSB recipes. 81.93 and 82.82 top-1 at 224x224.

  rwightman

  committed
  ddc29da

Commits on Nov 2, 2021

• Update eca_halonext26ts weights to a better set

  rwightman

  committed
  b328e56

• Better fix for huggingface#954 that doesn't break torchscript, pull torch._assert into timm namespace when it exists

  rwightman

  committed
  2ddef94

• Fix huggingface#954 by bringing traceable _assert into timm to allow compat w/ PyTorch < 1.8

  rwightman

  committed
  4f0f9cb

Commits on Oct 29, 2021

• Add interpolation mode handling to transforms. Removes InterpolationMode warning. Works for torchvision versions w/ and w/o InterpolationMode enum. Fix huggingface#738.

  rwightman

  committed
  a41de1f

• Add repr to auto_augment and random_erasing impl

  rwightman

  committed
  ed41d32

Commits on Oct 28, 2021

• Fix sam result again for imagenetv2

  rwightman

  committed
  135a48d

• Add new 50ts attn models to benchmark/meta csv files

  rwightman

  committed
  aaff2d8

• Fixed botne*t26 model results, add some 50ts self-attn variants

  rwightman

  committed
  1e17863

• Add weights for lambda_resnet50ts, halo2botnet50ts, lamhalobotnet50ts, updated halonet50ts

  rwightman

  committed
  ae72d00

Commits on Oct 26, 2021

• Add benchmark and metadata csv files

  rwightman

  committed
  13178ba

Commits on Oct 25, 2021

• Fix formatting of last commit

  rwightman

  committed
  b745d30

• Traceability fix for vit models for some experiments

  rwightman

  committed
  3478f1d

• Cleanup re-use of Dropout modules in Mlp modules after some twitter feedback :p

  rwightman

  committed
  f658a72

• Don't run profile if model is torchscripted

  rwightman

  committed
  71f00bf

• Merge pull request huggingface#933 from t-vi/unbind

  Show description for 7da1b0b
  rwightman

  authored
  7da1b0b

Commits on Oct 24, 2021

• Add activation count to fvcore based profiling in benchmark.py

  rwightman

  committed
  5882e62

• Update results.csv files with latest weights

  rwightman

  committed
  51f488b

• use .unbind instead of explicitly listing the indices

  t-vi

  committed
  f805ba8

• Fix some formatting in utils/model.py

  rwightman

  committed
  5799250

• add d8 and e8 regnetz models with group size 8

  rwightman

  committed
  0fe4fd3

Commits on Oct 21, 2021

• Update broken resnetv2_50 weight url, add resnetv1_101 a1h recipe weights for 224x224 train

  rwightman

  committed
  25e7c8c

Commits on Oct 20, 2021

• Support either deepspeed or fvcore for flop profiling

  rwightman

  committed
  f7325c7

Commits on Oct 19, 2021

• Add --bench profile mode for benchmark.py to just run deepspeed detailed profile on model

  rwightman

  committed
  6625379

• Add train size override and deepspeed GMACs counter (if deepspeed installed) to benchmark.py

  rwightman

  committed
  13a8bf7

• Update README.md

  rwightman

  authored
  0ba73e6

• Fixed eca_botnext26ts_256 weights added, 79.27

  rwightman

  committed
  b6caa35

• Add weights for regnetz_d and haloregnetz_c, update regnetz_c weights. Add commented PyTorch XLA code for halo attention

  rwightman

  committed
  c02334d

Commits on Oct 12, 2021

• Add option to include relative pos embedding in the attention scaling as per references. See discussion huggingface#912

  rwightman

  committed
  02daf2a

• Merge pull request huggingface#913 from ground0state/master

  Show description for 2c33ca6
  rwightman

  authored
  2c33ca6

• Fix bugs that Mixup does not work device=cpu

  ground0state

  committed
  047a5ec

• Remove some outdated comments, botnet networks working great now.

  rwightman

  committed
  cd34913

Commits on Oct 10, 2021

• Update lambda_resnet26t weights with better set

  rwightman

  committed
  6ed4cdc

• Merge pull request huggingface#910 from tmp-iclr/master

  Show description for 288ece0
  rwightman

  authored
  288ece0

• Add ConvMixer

  

  ICLR Author

  committed
  44d6d51

Pagination

PreviousNext