Nemotron Labs has introduced Audex (Nemotron-Labs-Audex-30B-A3B), a unified audio-text large language model. Built upon the Nemotron-Cascade-2-30B-A3B text-only model, Audex uses a single Transformer decoder to process and generate both audio and text seamlessly. The model was trained on a large dataset of 157.4 billion audio tokens and 320.5 billion text tokens, incorporating supervised training and reinforcement learning techniques. Audex achieves state-of-the-art performance in various audio tasks, including understanding, speech recognition, translation, and generation, while retaining the strong reasoning and knowledge capabilities of its text-based predecessor. AI
IMPACT This unified model architecture could advance multimodal AI capabilities, enabling more seamless integration of audio and text processing in future applications.
RANK_REASON The cluster describes a research paper detailing a new model release.
Read on Hugging Face Daily Papers →
- alphaXiv
- arXiv
- Audex
- Cascade RL
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- Nemotron-Cascade-2-30B-A3B
- Nemotron-Labs-Audex-30B-A3B
- ScienceCast
- Transformer++
AI-generated summary · Google Gemini · from 4 sources. How we write summaries →
