# v1.1.0

v1.1.0

• 1576fd47 Added RNA v5.3.0 models
• 0404594c Added --no-compile option to training
• 58e57fee Set training seed to random by default

v1.0.1

v1.0.1

• 5fed0a48 Fixes #426 where model was compiled twice when loading from checkpoint
• 41618b26 Bump torch to v2.9.1

v1.0.0

• 4df274e
  • Update dependencies to support the latest python (3.10-3.14), torch (2.9), cuda (12.8, 13.0) versions.
  • Update build-pipeline to pyproject style
  • Remove the dependency on FlashAttention
  • Torch does not yet support torch.compile on py3.14+ and training performance will be degraded, see pytorch#156856.
• f5db9e6 (f5db9e6c): Only autocast in the forward pass during training
• 1671d49 support removed for fast5.

v0.9.1

v0.9.1

• 4bfce35 v5.2 basecall models for DNA + RNA

v0.9.0

v0.9.0

• 205c5aa3: Updated dependencies and provide support for new python and pytorch versions.
• 84ebf8ed: Updated documentation on model training process and training data loading.
• 47690c54, 8144ddb0: Refactored training data loading process.
• 9573d7f0: Improved output from bonito evaluate to assist with debugging model development
• 4db046e1: Added support for dynamically loading different optimisers.
• 7261dbe3: Added RNA v5.1.0 models.
• aa5bc804: Remove mod-calling options from bonito. Please see dorado for supported mods and remora for modified basecall model development.

v0.8.1

• e33a860 Attention Is All You Need!
• 454324a v5.0.0 models and example training sets for DNA & RNA.
  • dna_r10.4.1_e8.2_400bps_sup@v5.0.0
  • dna_r10.4.1_e8.2_400bps_hac@v5.0.0
  • dna_r10.4.1_e8.2_400bps_fast@v5.0.0
  • rna004_130bps_sup@v5.0.0
  • rna004_130bps_hac@v5.0.0
  • rna004_130bps_fast@v5.0.0
  • example_data_dna_r10.4.1_v0
  • example_data_rna004_v0
• ed36968 new model configs.
• 37a0557 default alignment preset is now lr:hq.
• d4f6dd2 unpin bonito requirements.
• 40a9753 fast5 deprecation warning.
• 0ba190f fixed progress count when setting --max-reads.
• 6c8ecb5 batchnorm fusing for inference.
• 302b1ce bonito view now accepts a model directory or a config.
• a170b7c default scaling fixes.

v0.7.3

• 775ce75 v4.3 simplex models.
• 214bac6 RNA004 models.
• e87b05e RNA Remora fix.
• fc1d62a Add autocast to validation.

v0.7.2

• 0061534 Upgrade remora to 2.1.3.
• cd62fbb multi-channel input support.
• 7f79a08 Allow bonito to be run as a module.
• f403b40 Set v4 flag for bonito export.
• a9823b3 fix bonito convert nep-0034.

v0.7.1

Highlights

• 9113e24 v4.2.0 5kHz simplex models.
  • dna_r10.4.1_e8.2_400bps_fast@v4.2.0
  • dna_r10.4.1_e8.2_400bps_hac@v4.2.0
  • dna_r10.4.1_e8.2_400bps_sup@v4.2.0
• 8c96eb8 make sample_id optional for fast5 input.
• 3b4bcad ensure decoder runs on same device as nn model.
• 8fe1f61 fix training data downloading.
• 26d52d9 set default --valid-chunks to None.
• ebc32a0 fix models as list.

Thanks @chAwater for his collection of bug fixes in this release.

Installation

$ pip install ont-bonito

Note: For anything other than basecaller training or method development please use dorado.

v0.7.0

Highlights

• 66ee29a v4.1.0 simplex models.
  • dna_r10.4.1_e8.2_260bps_fast@v4.1.0
  • dna_r10.4.1_e8.2_260bps_hac@v4.1.0
  • dna_r10.4.1_e8.2_260bps_sup@v4.1.0
  • dna_r10.4.1_e8.2_400bps_fast@v4.1.0
  • dna_r10.4.1_e8.2_400bps_hac@v4.1.0
  • dna_r10.4.1_e8.2_400bps_sup@v4.1.0
• 4cf3c6f torch 2.0 + updated requirements.
• 3bc338a fix use of TLEN.
• 21df7d5 v4.0.0 simplex models.
  • dna_r10.4.1_e8.2_260bps_fast@v4.0.0
  • dna_r10.4.1_e8.2_260bps_hac@v4.0.0
  • dna_r10.4.1_e8.2_260bps_sup@v4.0.0
  • dna_r10.4.1_e8.2_400bps_fast@v4.0.0
  • dna_r10.4.1_e8.2_400bps_hac@v4.0.0
  • dna_r10.4.1_e8.2_400bps_sup@v4.0.0

Installation

Torch 2.0 (from pypi.org) is now built using CUDA 11.7 so the default installation of ont-bonito can be used for Turing/Ampere GPUs.

$ pip install ont-bonito

Note: For anything other than basecaller training or method development please use dorado.