What's Changed
- Fix XDG_CACHE_HOME double .cache in path construction by @jnMetaCode in #1828
- fix(llamacpp): correct EOS attention mask and vocab truncation bugs by @giulio-leone in #1820
- fix(dsl): add JSON double-quotes to string literals inside container types by @giulio-leone in #1821
- fix: accumulate duplicate token IDs in vocabulary construction by @RoloPineda in #1831
- fix: inline SPIECE_UNDERLINE fallback for transformers compat by @giulio-leone in #1832
- fix: preserve all token IDs when multiple BPE tokens decode to same string by @alvinttang in #1833
- Fix documentation preview action + update uv.lock by @RobinPicard in #1838
- Update the Dottxt model by @RobinPicard in #1845
- Pin the version of dottxt to >=0.2 in the pyproject by @RobinPicard in #1857
- Refactor _check_chat_template method to improve tokenizer compatibility checks by @Ki-Seki in #1855
- Handle both HF tokenizer and mlx tokenizer wrapper for the MLXLM model by @RobinPicard in #1848
New Contributors
- @jnMetaCode made their first contribution in #1828
- @giulio-leone made their first contribution in #1820
- @RoloPineda made their first contribution in #1831
- @alvinttang made their first contribution in #1833
Full Changelog: 1.2.12...1.2.13