AI BriefingNVIDIA Releases Nemotron-Labs-Dory1.0-4B, a Looped Latent-Recurrent LLM with 256K Context
Key point
The 4B parameter model uses a looped architecture with interleaved sliding-window and full attention to scale intelligence on a small footprint.
Details
NVIDIA released NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16, a 4B parameter large language model trained with a looped (latent-recurrent) architecture. The model employs a normalized-GPT structure with interleaved sliding-window attention (SWA) and full-attention layers, allowing it to apply more compute per token to scale intelligence within a small footprint.
Key Specifications
- Architecture: Looped (latent-recurrent) normalized-GPT with interleaved SWA and full-attention layers.
- Parameters: 4B.
- Precision: BF16.
- Context Length: Up to 256K tokens.
- Languages: English.
- License: OpenMDW License Agreement, version 1.1.
- Release Date: October 16, 2026.
The model is designed for general-purpose reasoning and is available for both commercial and non-commercial use.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Black Forest Labs Releases FLUX 3 Image Klein: 5.6B Open Weights Model for Text-to-Image and Editing