Arthur

[`config.to_dict()`] update test and models that have a composition

3

# What does this PR do? Adresses #23876 Most of the configurations that are `composition` are not json serialzable. - A test exists, but is usually not run (`ConfigTester` scarcly...

Add training support for EnCodec

15

### Feature request Would be cool to add training support for the EnCodec model. Not entirely sure if we can easily make it compatible with Trainer, so this can be...

[`Core tokenization`] `add_dummy_prefix_space` option to help with latest issues

5

# What does this PR do? Allows users to use `tokenizer.tokenize` controlling the addition of prefix space. Let's also update fast!

[`Add Mamba`] Adds support for the `Mamba` models

6

# What does this PR do? - [x] Implement cpu ops - [x] Add integration tests - [ ] Implement fast path - [ ] check training + peft -...

[`Do not Merge`]

1

# What does this PR do? UV

[`remove black`] And use ruff

1

To apply

[ `TokenizationLlama`] fix the way we convert tokens to strings to keep leading spaces

1

# What does this PR do? Fixes #29452

# What does this PR do? Clean the code, similar to #29492

[`TokenizationUtils`] add support for `split_special_tokens`

1

# What does this PR do? Adds support for `split_special_tokens` for fast models as well - [ ] deprecate `split_special_tokens` for `encode_special_tokens` for API consistency - [ ] make sure...

[`LlamaSlowConverter`] Slow to Fast better support

1

# What does this PR do? Makes sur that the eos token, bos token and unk token are correctly taken when initializing from slow. Fixes #29774