Hugging Face Transformers: Patch release v5.22.1
Hugging Face Transformers v5.22.1 is a minor documentation-only patch release. It fixes an issue by removing a work-in-progress document for Ml4.
transformers release archive — Page 1
5 tracked entries. Older history is imported progressively; see source coverage.
Hugging Face Transformers: v5.22.0: Mistral Large 4 (Le Chonk)
Hugging Face Transformers v5.22.0 introduces the Mistral Large 4 1T-parameter model and adds support for FP8 adapter training. The release includes a breaking architectural change regarding Execution Provider (EP) logic separation.
Hugging Face Transformers: Patch release v5.21.1
Hugging Face Transformers v5.21.1 is a patch release that fixes the rope_scaling setter for configurations that assign it during initialization. This update resolves a critical regression that broke remote code loading in vLLM.
Hugging Face Transformers: Release: v5.21.0
Hugging Face Transformers v5.21 introduces support for multiple new models (such as MiniCPM-OCR, Flint, UltraVBert, and Gradium/Pocket TTS) and enforces breaking architectural changes including custom tensor parallelism and MoE expert sharding. Developers upgrading to this version should review updated TP plans and custom MoE implementations.
Hugging Face Transformers: Patch release v5.20.2
Hugging Face Transformers v5.20.2 is a patch release addressing critical bugs in multimodal generation inputs, bfloat16 expert parallelism gradients, and eager_mask host-to-device copies. It ensures correct input typing within `prepare_inputs_for_generation` and prevents performance regressions in masking.