transformers Release v5.23.0

Hugging Face Transformers v5.23.0 introduces three new models: Inkling2 Small, a multimodal model with a 594B parameter mixture-of-experts architecture and image-to-text capabilities; JEV, an experimental decision head model for scoring single-token outputs with a soft-capped linear layer; and LFM3, a hybrid Liquid Foundation Model using gated convolutions and grouped-query attention with sandwich normalization. These additions expand the library's support for advanced multimodal, decision-making, and efficient inference architectures.