The Hub is quiet on big lab drops. Hugging Face put a salary behind Apple Silicon local inference this morning. Hugging Face also shipped a tokenizer rewrite that keeps the same IDs. Xiaomi added a MIT omnimodal checkpoint to the Hub yesterday.
On the Hub
Jun Kim joins Hugging Face to keep oMLX on Apple Silicon
Hugging Face said Jun Kim is joining full time to support the MLX community.1 Kim created and maintains oMLX. The project stays Apache 2.0. The move puts a team behind Apple Silicon local inference.
tokenizers v1 ships with same IDs and a faster Rust core
Hugging Face released tokenizers v1 on September 21.2 The library keeps the same token IDs, API and vocabulary as v0.23. The blog says the rewrite is often by tens of times faster. It is infrastructure, not a model.
Xiaomi MiMo-V2.6-Flash-RL lands on the Hub
XiaomiMiMo added MiMo-V2.6-Flash-RL to the Hub on September 21.3 The card is MIT licensed. The Hub lists about 159.4B of safetensors. The card claims native text, image, video and audio with a 1M context. A sibling Pro-RL appeared the same morning.
