InjazAI

Hugging Face Transformers 5.18.0 adds three notable models: Nemotron 3 Diarization for streaming speaker diarization, NVIDIA's NemotronH Omni multimodal reasoner, and NAVER's HyperCLOVAX Vision V2, expanding open-model options for MENA developers.

1 min read

Transformers 5.18.0 Release Adds Nemotron 3 Diarization, NemotronH Omni, HyperCLOVAX Vision V2

What's new in Transformers 5.18.0?

FAQ

What is Nemotron 3 Diarization?

It is an open-weight speaker diarization model that determines who spoke when in real-world audio, supporting both streaming and offline inference for up to eight speakers.

How does NemotronH Omni differ from other multimodal models?

NemotronH Omni pairs a RADIO vision encoder with an optional Parakeet-based sound encoder, using a single autoregressive model to reason jointly over text, images, video, and audio.

Does this release matter for MENA teams?

Yes. Speaker diarization helps analyze Arabic call-center audio, multilingual GTE supports Arabic semantic search, and NemotronH Omni benefits video and audio analytics use cases.

What other major changes ship in 5.18.0?

The release includes broad MoE and cache fixes, GGUF support for MoE, video processing improvements, and an update to torch 2.14.

Source: Hugging Face Transformers Releases

AI-assisted summary of the source above. High-confidence stories publish automatically; others wait for review. Facts that are not in the source are not added.