## [0.7.0] - Nov 4th, 2025 - `RecipeModel` is now callable and forwards the call to `RecipeModel.module` for a cleaner, more convenient syntax. - A new `get_asset_download_manager` helper function to download assets in procedural code. - A new `register_recipe_assets` helper function that can be used to register recipe-specific asset cards that cannot be (accidentally) overwritten by users. [More info](https://github.com/facebookresearch/fairseq2/pull/1373) - Reference API documentation has been flattened and updated for better readability. - Revised Wav2Vec 2.0 recipes have been merged back and are available under the recipes/wav2vec2 directory.

- Fixes LLaMA 3 tokenizer load bug

Bug fixes

## What's Changed * Disable digital attestation check by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/942 * enhance unit tests for position encoder by @zyaoj in https://github.com/facebookresearch/fairseq2/pull/940 * Bump to v0.4.0dev0 by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/943 * Fix error wrongly triggered on 'model' key absence in Llama checkpoint conversion by @MartinGleize in https://github.com/facebookresearch/fairseq2/pull/944 * Make Jepa loader more flexible by @antoine-tran in https://github.com/facebookresearch/fairseq2/pull/945 * Introduce RuntimeContext by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/946 * Revise library setup by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/947 * Introduce SequenceGeneratorHandler by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/948 * Improve CLI by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/949 * Nit updates by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/950 * Move EarlyStopper to recipes by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/951 * Move console accessors to recipes.utils.rich by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/952 * Revise fairseq2.optim by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/953 * Revise LR schedulers by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/954 * Revise datasets by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/955 * Refactor dataset configs by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/956 * Extend VocabularyInfo dataclass by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/957 * Add script to write Llama's HF-formatted config.json for vLLM by @MartinGleize in https://github.com/facebookresearch/fairseq2/pull/936 * Remove setup_fairseq2 from HF config recipe by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/959 * Training DPO with pre-computed reference scores by @uralik in https://github.com/facebookresearch/fairseq2/pull/958 * Fix the dev tag of the PEP 440 version by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/962 * Fix dump-config bug by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/961 * Fix config_kls bug in SamplerHandlers by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/963 * Refactor model configs by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/965 * Move batching strategy to DataReadConfig by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/968 * Nit updates by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/972 * Fix LLaMA checkpoint by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/973 * Enforce right package import paths by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/974 * Allow training checkpoints to contain rich objects by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/978 * Support runtime context in recipe loaders by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/979 * Refactors first party recipes by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/984 * Fix intermediate_size calculation in Llama config convert function by @MartinGleize in https://github.com/facebookresearch/fairseq2/pull/982 * Fix LLaMA test by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/985 * Introduce to_gangs helper by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/986 * Use logprob scores in sampling generator by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/987 * Introduce abstract ASR model and revise eval recipe by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/988 * Improve best checkpoint handling by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/989 * Improve IO error handling by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/990 * Generate Hugging Face config.json by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/991 * Fix CheckpointManager bugs by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/994 * using dtype and device in hub.load by @artemru in https://github.com/facebookresearch/fairseq2/pull/995 * Last refactoring bundle for 0.4 by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/998 * doc update sprint 4 by @zyaoj in https://github.com/facebookresearch/fairseq2/pull/993 * Add dataset extras option to recipes by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/999 * Support PyTorch 2.6 by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/1000 * Fix PT2.6 linting issues by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/1002 * Bump to v0.4.0 by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/1001 **Full Changelog**: https://github.com/facebookresearch/fairseq2/compare/v0.3.0...v0.4.0

See [changelog](CHANGELOG.md). ## What's Changed * Make tbb optional by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/4 * Call non-typed Storage.data instead of unsafe_data by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/5 * Remove tbb link when not used by @vanclevstik in https://github.com/facebookresearch/fairseq2/pull/8 * Improve CI pipeline by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/11 * Support for integer sample_rate by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/12 * Support for Homebrew on ARM64 by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/13 * Add legal links to the doc by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/14 * Nit CI improvements by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/17 * Improve doc build by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/18 * Update markdown docs by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/21 * Dump to v0.1.1 by @cbalioglu in https://github.com/facebookresearch/fairseq2/pull/22 **Full Changelog**: https://github.com/facebookresearch/fairseq2/compare/v0.1.0...v0.1.1