Hugging Face Transformers v5.14.0 Released with Inkling Multimodal Model (975B Parameters)

It is a new release of the Transformers library adding the Inkling multimodal model from Thinking Machines (975B total, 41B active) and major Flash Attention performance improvements and MTP support.

A general-purpose multimodal model from Thinking Machines with 975B total parameters (41B active) accepting text, image, and audio inputs, suitable for enterprise applications in the region.

It provides up to 260% performance improvements and powerful multimodal models, reducing infrastructure costs and accelerating deployment in enterprise and government applications.

GPTNeoX now remaps embed_out to lm_head, and GPTBigCode has _supports_attention_backend = True enabled for vLLM compatibility, requiring code updates for existing users.