Code

Add TimmWrapper#34564

Code

Merged

qubvel merged 92 commits intohuggingface:mainhuggingface/transformers:mainfrom qubvel:timm-wrapperqubvel/transformers:timm-wrapperCopy head branch name to clipboardDec 11, 2024

+1,432-129

Commits

Commits on Nov 1, 2024

• Add files

  

  amyeroberts

  authored and

  qubvel

  committed
  39c42db

  Copy full SHA for 39c42db

• Init

  

  amyeroberts

  authored and

  qubvel

  committed
  4b98375

  Copy full SHA for 4b98375

• Add TimmWrapperModel

  

  amyeroberts

  authored and

  qubvel

  committed
  aa494f1

  Copy full SHA for aa494f1

• Fix up

  

  amyeroberts

  authored and

  qubvel

  committed
  44d123e

  Copy full SHA for 44d123e

• Some fixes

  

  amyeroberts

  authored and

  qubvel

  committed
  4b35ae2

  Copy full SHA for 4b35ae2

• Fix up

  

  amyeroberts

  authored and

  qubvel

  committed
  2b5db8f

  Copy full SHA for 2b5db8f

• Remove old file

  

  amyeroberts

  authored and

  qubvel

  committed
  b07a5c9

  Copy full SHA for b07a5c9

• Sort out import orders

  

  amyeroberts

  authored and

  qubvel

  committed
  e3a88b6

  Copy full SHA for e3a88b6

• Fix some model loading

  

  amyeroberts

  authored and

  qubvel

  committed
  baffbe2

  Copy full SHA for baffbe2

• Compatible with pipeline and trainer

  

  amyeroberts

  authored and

  qubvel

  committed
  6fc50cf

  Copy full SHA for 6fc50cf

• Fix up

  

  amyeroberts

  authored and

  qubvel

  committed
  ed00b41

  Copy full SHA for ed00b41

• Delete test_timm_model_1/config.json

  

  amyeroberts

  authored and

  qubvel

  committed
  50b507b

  Copy full SHA for 50b507b

• Remove accidentally commited files

  

  amyeroberts

  authored and

  qubvel

  committed
  de87e54

  Copy full SHA for de87e54

• Delete src/transformers/models/modeling_timm_wrapper.py

  

  amyeroberts

  authored and

  qubvel

  committed
  a907061

  Copy full SHA for a907061

• Remove empty imports; fix transformations applied

  

  amyeroberts

  authored and

  qubvel

  committed
  1f55841

  Copy full SHA for 1f55841

• Tidy up

  

  amyeroberts

  authored and

  qubvel

  committed
  23b38af

  Copy full SHA for 23b38af

• Add image classifcation model to special cases

  

  amyeroberts

  authored and

  qubvel

  committed
  2f0aee1

  Copy full SHA for 2f0aee1

• Create pretrained model; enable device_map='auto'

  

  amyeroberts

  authored and

  qubvel

  committed
  bff2e98

  Copy full SHA for bff2e98

• Enable most tests; fix init order

  

  amyeroberts

  authored and

  qubvel

  committed
  0c80253

  Copy full SHA for 0c80253

• Sort imports

  

  amyeroberts

  authored and

  qubvel

  committed
  a32de6a

  Copy full SHA for a32de6a

• [run-slow] timm_wrapper

  

  amyeroberts

  authored and

  qubvel

  committed
  59f55d1

  Copy full SHA for 59f55d1

• Pass num_classes into timm.create_model

  qubvel

  committed
  666419f

  Copy full SHA for 666419f

• Remove train transforms from image processor

  qubvel

  committed
  5540d32

  Copy full SHA for 5540d32

• Update timm creation with pretrained=False

  qubvel

  committed
  88f737c

  Copy full SHA for 88f737c

• Fix gamma/beta issue for timm models

  qubvel

  committed
  496d38d

  Copy full SHA for 496d38d

• Fixing gamma and beta renaming for timm models

  qubvel

  committed
  4cfa51f

  Copy full SHA for 4cfa51f

• Simplify config and model creation

  qubvel

  committed
  3891975

  Copy full SHA for 3891975

• Remove attn_implementation diff

  qubvel

  committed
  f878ab4

  Copy full SHA for f878ab4

• Fixup

  qubvel

  committed
  b1f145f

  Copy full SHA for b1f145f

• Docstrings

  qubvel

  committed
  7d87a95

  Copy full SHA for 7d87a95

• Fix warning msg text according to test case

  qubvel

  committed
  ae0425d

  Copy full SHA for ae0425d

Commits on Nov 4, 2024

• Fix device_map auto

  qubvel

  committed
  ec9eade

  Copy full SHA for ec9eade

• Set dtype and device for pixel_values in forward

  qubvel

  committed
  9767048

  Copy full SHA for 9767048

• Enable output hidden states

  qubvel

  committed
  d5478f6

  Copy full SHA for d5478f6

• Enable tests for hidden_states and model parallel

  qubvel

  committed
  b61048b

  Copy full SHA for b61048b

• Remove default scriptable arg

  qubvel

  committed
  72236b5

  Copy full SHA for 72236b5

• Refactor inner model

  qubvel

  committed
  e290f22

  Copy full SHA for e290f22

• Update timm version

  qubvel

  committed
  0a08a1f

  Copy full SHA for 0a08a1f

Commits on Nov 5, 2024

• Fix _find_mismatched_keys function

  qubvel

  committed
  1faf9ec

  Copy full SHA for 1faf9ec

• Change inheritance for Classification model (fix weights loading with device_map)

  qubvel

  committed
  addbac8

  Copy full SHA for addbac8

Commits on Nov 19, 2024

• Minor bugfix

  qubvel

  committed
  8ba4592

  Copy full SHA for 8ba4592

• Disable save pretrained for image processor

  qubvel

  committed
  2b993a7

  Copy full SHA for 2b993a7

• Rename hook method for loaded keys correction

  qubvel

  committed
  0b00cef

  Copy full SHA for 0b00cef

• Rename state dict keys on save, remove timm_model prefix, make checkpoint compatible with timm

  qubvel

  committed
  485fe7a

  Copy full SHA for 485fe7a

• Managing num_labels <-> num_classes attributes

  qubvel

  committed
  dede4e4

  Copy full SHA for dede4e4

• Enable loading checkpoints in Trainer to resume training

  qubvel

  committed
  1ec132c

  Copy full SHA for 1ec132c

Commits on Nov 20, 2024

• Update error message for output_hidden_states

  qubvel

  committed
  d96b257

  Copy full SHA for d96b257

• Add output hidden states test

  qubvel

  committed
  6b1b621

  Copy full SHA for 6b1b621

• Decouple base and classification models

  qubvel

  committed
  bb7d465

  Copy full SHA for bb7d465

• Add more test cases

  qubvel

  committed
  4390213

  Copy full SHA for 4390213

• Add save-load-to-timm test

  qubvel

  committed
  4233850

  Copy full SHA for 4233850

• Merge branch 'main' into timm-wrapper

  qubvel

  authored
  ef96b6d

  Copy full SHA for ef96b6d

• Fix test name

  qubvel

  committed
  c3bb39a

  Copy full SHA for c3bb39a

• Fixup

  qubvel

  committed
  5444a6c

  Copy full SHA for 5444a6c

Commits on Nov 27, 2024

• Add do_pooling

  qubvel

  committed
  06039d0

  Copy full SHA for 06039d0

• Add test for do_pooling

  qubvel

  committed
  0b78735

  Copy full SHA for 0b78735

• Fix doc

  qubvel

  committed
  5bb2950

  Copy full SHA for 5bb2950

• Add tests for TimmWrapperModel

  qubvel

  committed
  38b9423

  Copy full SHA for 38b9423

• Add validation for num_classes=0 in timm config + test for DINO checkpoint

  qubvel

  committed
  c1cc1fa

  Copy full SHA for c1cc1fa

• Adjust atol for test

  qubvel

  committed
  3acafad

  Copy full SHA for 3acafad

• Fix docs

  qubvel

  committed
  8e8f41b

  Copy full SHA for 8e8f41b

Commits on Nov 28, 2024

• dev-ci

  qubvel

  committed
  c5c288f

  Copy full SHA for c5c288f

• dev-ci

  qubvel

  committed
  a4ae76f

  Copy full SHA for a4ae76f

• Add tests for image processor

  qubvel

  committed
  464874f

  Copy full SHA for 464874f

• Update docs

  qubvel

  committed
  02fbd58

  Copy full SHA for 02fbd58

• Update init to new format

  qubvel

  committed
  eb9a66f

  Copy full SHA for eb9a66f

• Update docs in configuration

  qubvel

  committed
  e08fe70

  Copy full SHA for e08fe70

• Fix some docs in image processor

  qubvel

  committed
  7734804

  Copy full SHA for 7734804

• Improve docs for modeling

  qubvel

  committed
  051acee

  Copy full SHA for 051acee

• fix for is_timm_checkpoint

  qubvel

  committed
  185fab8

  Copy full SHA for 185fab8

• Update code examples

  qubvel

  committed
  bca3279

  Copy full SHA for bca3279

• Fix header

  qubvel

  committed
  672cc6d

  Copy full SHA for 672cc6d

• Fix typehint

  qubvel

  committed
  983b9b2

  Copy full SHA for 983b9b2

• Increase tolerance a bit

  qubvel

  committed
  9c128ce

  Copy full SHA for 9c128ce

Commits on Nov 29, 2024

• Merge branch 'main' into timm-wrapper

  qubvel

  authored
  f2dba79

  Copy full SHA for f2dba79

• Fix Path

  qubvel

  committed
  9bc887b

  Copy full SHA for 9bc887b

Commits on Dec 2, 2024

• Fixing model parallel tests

  qubvel

  committed
  f92216f

  Copy full SHA for f92216f

• Disable "parallel" tests

  qubvel

  committed
  42278a7

  Copy full SHA for 42278a7

• Add comment for metadata

  qubvel

  committed
  6b3ba3b

  Copy full SHA for 6b3ba3b

• Refactor AutoImageProcessor for timm wrapper loading

  qubvel

  committed
  ff6efde

  Copy full SHA for ff6efde

• Remove custom test_model_outputs_equivalence

  qubvel

  committed
  a476610

  Copy full SHA for a476610

• Add require_timm decorator

  qubvel

  committed
  327095a

  Copy full SHA for 327095a

• Fix comment

  qubvel

  committed
  7e0d2c6

  Copy full SHA for 7e0d2c6

Commits on Dec 3, 2024

• Make image processor work with older timm versions and tensor input

  qubvel

  committed
  a10bc0d

  Copy full SHA for a10bc0d

• Save config instead of whole model in image processor tests

  qubvel

  committed
  a87fbf6

  Copy full SHA for a87fbf6

• Add docstring for image_processor_filename

  qubvel

  committed
  90c1c88

  Copy full SHA for 90c1c88

• Sanitize kwargs for timm image processor

  qubvel

  committed
  fd7b646

  Copy full SHA for fd7b646

• Fix doc style

  qubvel

  committed
  cdd3811

  Copy full SHA for cdd3811

• Update check for tensor input

  qubvel

  committed
  3d1a76e

  Copy full SHA for 3d1a76e

Commits on Dec 5, 2024

• Merge branch 'main' into timm-wrapper

  qubvel

  authored
  cc0a330

  Copy full SHA for cc0a330

Commits on Dec 11, 2024

• Update normalize

  qubvel

  committed
  4edfe90

  Copy full SHA for 4edfe90

• Remove _load_timm_model function

  qubvel

  committed
  8b82e2b

  Copy full SHA for 8b82e2b