NVIDIA is also releasing new multi-speaker speech AI models, a new model with reasoning capabilities and datasets for AI safety, as well as open tools to generate high-quality synthetic datasets for reinforcement learning and domain-specific model customization. These tools include:
MultiTalker Parakeet : An automatic speech recognition model for streaming audio that can understand multiple speakers, even in overlapped or fast-paced conversations.
Sortformer : A state-of-the-art model that can accurately distinguish multiple speakers within an audio stream — a process called diarization — in real time.
Nemotron Content Safety Reasoning : A reasoning-based AI safety model that dynamically enforces custom policies across domains.
Nemotron Content Safety Audio Dataset : A synthetic dataset that helps train models to detect unsafe audio content, enabling the development of guardrails that work across text and audio modalities.
NeMo Gym : an open-source library that accelerates and simplifies the development of reinforcement learning environments for LLM training. NeMo Gym also contains a growing collection of ready-to-use training environments to enable Reinforcement Learning from Verifiable Reward (RLVR).
NeMo Data Designer Library : Now open-sourced under Apache 2.0, this library provides an end-to-end toolkit to generate, validate and refine high-quality synthetic datasets for generative AI development, including domain-specific model customization and evaluation.
NVIDIA ecosystem partners using NVIDIA Nemotron and NeMo tools to build secure, specialized agentic AI include CrowdStrike, Palantir and ServiceNow.
NeurIPS attendees can explore these innovations at the Nemotron Summit , taking place today, from 4-8 p.m. PT, with an opening address by Bryan Catanzaro, vice president of applied deep learning research at NVIDIA.
Of the dozens of NVIDIA-authored research papers at NeurIPS , here are a few highlights advancing language models:
Audio Flamingo 3: Advancing Audio Intelligence With Fully Open Large Audio Language Models : This large audio language model is capable of reasoning across speech, sound and music. It can understand and reason audio segments up to 10 minutes in length, achieving state-of-the-art results on over 20 benchmarks.
Minitron-SSM: Efficient Hybrid Language Model Compression Through Group-Aware SSM Pruning : This poster introduces a pruning method capable of compressing hybrid models, demonstrated by pruning and distilling Nemotron-H 8B from 8 billion to 4 billion parameters. The resulting model surpasses the accuracy of similarly sized models while achieving 2x faster inference throughput.
Jet-Nemotron: Efficient Language Model With Post Neural Architecture Search : This work presents a cost-efficient post-training pipeline for developing new efficient language model architectures, and introduces a hybrid-architecture model family produced with the pipeline. These models match or surpass the accuracy of leading full-attention baselines while delivering substantially higher generation throughput.
Nemotron-Flash: Towards Latency-Optimal Hybrid Small Language Models : This project introduces a new small language model (SLM) architecture that redesigns SLMs around real-world latency rather than parameter count — achieving state-of-the-art speed and accuracy.
ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models : Prolonged reinforcement learning, or ProRL, is a technique that extends model training over longer periods. In this NeurIPS poster, NVIDIA researchers describe how this methodology results in models that consistently outperform base models for reasoning.
View the full list of events at NeurIPS , running through Sunday, Dec. 7, in San Diego.
See notice regarding software product information.
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://blogs.nvidia.com/blog/neurips-open-source-digital-physical-ai/#content
- https://www.nvidia.com/en-us/
- https://blogs.nvidia.com/?s=
- Asus swaps out the PCIe x16 connector for x8 on new RTX 5060 Ti GPUs — Gigabyte does the opposite with x16 upgrade to its WindForce Max card
- Robots’ Holiday Wishes Come True: NVIDIA Jetson Platform Offers High-Performance Edge AI at Festive Prices
- Sapphire PR manager wishes AMD and Nvidia would let partners run wild with design — wants freedom to bring back Toxic line more often
- The 'ExtrudeX' machine wants to turn your 3D printing waste into reusable filament, all at home — this Kickstarter project is itself 3D-printable with minimal h
- Framework raises DDR5 RAM upgrade prices by 50% amid DRAM shortage — only for Laptop DIY edition, says prices will likely rise again
Informational only. No financial advice. Do your own research.