Skip to content

CHANGELOG.md

Latest commit

 

History

History
112 lines (68 loc) · 2.52 KB

CHANGELOG.md

File metadata and controls

112 lines (68 loc) · 2.52 KB

Change Log

Unpublished

  • Added support for loading pytorch models from HF hub via URL prefix [hf-pytorch].

v0.2.14 - 2023-05-15

  • Added support for LoRA training.

v0.2.13 - 2023-05-03

  • Adding VGG models.
  • Fixing bugs in SAM model inference.

v0.2.12 - 2023-04-17

  • Fixing small bugs in SAM.

v0.2.11 - 2023-04-17

  • Added Segment Aything models.

v0.2.10 - 2023-02-27

  • Added post-pooling features for ConvNeXt in feature dictionary.

v0.2.9 - 2022-10-28

  • Exposed attention map in ViT models.

v0.2.8 - 2022-09-05

  • tfimm now supports python 3.10.

v0.2.7 - 2022-06-14

  • Added EfficinentNet and MobileNet-V2 models.

v0.2.6 - 2022-05-13

  • Added tiny and small ConvNeXt models.

v0.2.5 - 2022-02-21

  • Preprocessing works for abritrary number of in_channels.
  • Removed temporary version restriction on `libclang``

v0.2.4 - 2022-01-31

  • Adding PiT models
  • Simplified pretrained parameter in create_function.
  • Added model-specific cache
  • Added adaptability of in_channels

v0.2.3 - 2022-01-20

  • Added ConvNeXt models
  • Added PoolFormer models
  • Improved LR schedulers for training framework

v0.2.2. - 2022-01-17

  • Improvements to the training framework.

v0.2.1 - 2022-01-07

  • Small changes to the training framework.

v0.2.0 - 2022-01-02

  • Added hybrid Vision Transformers (vit_hybrid).
  • Added resnetv2 module, which inlcudes Big Transfer (BiT) resnets.
  • Added Pyramid Vision Transformer models
  • Added first version of training framework (tfimm/train). Still work in progress. Possibly buggy.

v0.1.5 - 2021-12-12

  • Added option for models to return intermediate features via return_features parameter
  • Added DropPath regularization to vit module (stochastic depth)
  • Added ability to load saved models from a local cache

v0.1.4 - 2021-12-08

  • Fixed bug with dropout in Classifier layer

v0.1.3 - 2021-12-07

  • Added CaiT models
  • Added MLP-Mixer, ResMLP and gMLP models
  • Added ResNet models
  • Fixed bug with Swin Transformer and mixed precision

v0.1.2 - 2021-11-25

  • Reduced TF version requirement from 2.5 to 2.4.
  • Added ConvMixer models
  • Added Swin Transformer models

v0.1.1 - 2021-11-22

  • Refactored code in resnet.py.
  • Added create_preprocessing function to generate model-specific preprocessing.
  • Added profiling results (max batch size and throughput for inference and backpropagation) for K80 and V100 (float32 and mixed precision) GPUs.
  • Fixed bug with ViT models and mixed precision.

v0.1.0 - 2021-11-17

  • First release.