The Tev1-4B-experimental weights are in the togethercomputer repository on Hugging Face. Together AI describes the model as an experiment meant to show how specialized decision models can be trained fairly easily.
Victor M says the model turns long documents into small page images to save tokens, then retrieves the full text of only the pages relevant to a question.
The 100-million-parameter model tracks who spoke when, including when voices overlap, and supports up to eight speakers. It is available on huggingface.
Nemotron 3 Diarization can be tried in a Hugging Face Space. A Hugging Face blog covers how the model works and how to get started.
Trained on Hugging Face Jobs for about $1.50, the model proposes a task tag and probability from a dataset's column names and first row. Daniel van Strien says 17 minutes of fine-tuning raised its first-suggestion match with an owner's tag from 10% to 69%.
Pedro Cuenca said tokenizers v1 (Rust) is 3 to 30 times faster, with an opt-out tokenization training crate, additional optimizations, and extensive benchmarks. He said it is coming soon to transformers.
Lysandre (@LysandreJik) said Tokenizers v1’s first release candidate is out, citing up to 30x faster tokenization, thread scaling, lower latency, lower memory use, and a very small crate size. He linked details at https://huggingface.co/blog/tokenizers-v1.
Daniel van Strien said robotics company Eidon AI is winding down and has released all its data on Hugging Face under CC-BY-4.0, including 1,274 hours of egocentric video with paired 7-IMU arm tracking across 13,451 recordings (9 TB).
Qwen highlighted a Hugging Apps live demo of Qwen-Image-2.1 on Hugging Face Spaces, saying one checkpoint covers generation and editing and can be tried in the browser with no setup. Hugging Apps described the model as a 7B-parameter native generation and editing model with up to 10 image references.
Ghita told Jev users that zerank-2 is a 4B open-source reranker that also returns a confidence score, so a threshold can be set to use it as a classifier. The post said it can be self-hosted with Baseten or Fireworks AI and linked a Hugging Face model page.
Hugging Apps says Qwen Image 2.1 is a 7B-parameter native image generation and editing model with up to 10 image references, its own prompt enhancement LLMs, and integrations with diffusers and ComfyUI. It is available on Hugging Face Spaces.
Sayak Paul said Qwen-Image-2.1 is available Day-0 in Diffusers and linked to the model on Hugging Face.
