open source

Nemotron-4 15B

Provided by: Framework: Unknown

Nemotron-4 15B is a 15-billion-parameter multilingual large language model developed by NVIDIA as part of the NeMo framework. Trained on over 8 trillion tokens, it delivers strong performance in reasoning, multilingual generation, and code understanding tasks. Built with a transformer-based architecture using grouped-query attention and rotary embeddings, Nemotron-4 15B achieves competitive results across major benchmarks, outperforming similar open-weight models. Designed for research and enterprise use, it enables efficient deployment on NVIDIA GPUs through the NeMo and TensorRT-LLM ecosystems, supporting scalable inference for diverse AI applications.

Model Performance Statistics

0

Views

January 30, 2025

Released

Aug 19, 2025

Last Checked

4.0

Version

Capabilities
  • Synthetic data generation
  • Instruction tuning
  • Technical QA
Performance Benchmarks
diversity92%
factuality89%
Technical Specifications
Parameter Count
N/A
Training & Dataset

Dataset Used

Technical documentation, textbooks

Related AI Models

Discover similar AI models that might interest you

Modelopen source

Orca 2 13B

Orca 2 13B

Orca 2 13B

Microsoft

Orca 2.13 B is a large language model developed by Microsoft Research to enhance reasoning and comprehension in smaller models. Built on top of Meta’s LLaMA 2 architecture, it utilizes synthetic training data to simulate advanced reasoning strategies, including step-by-step deduction and self-reflection. Orca 2 demonstrates strong performance in logic, math, and reading comprehension, closing the gap between smaller open models and much larger proprietary systems. It serves as an open research model for studying how efficient LLMs can reason with minimal computational resources.

Natural Language Processingai-modelsAI research model
0
Modelopen source

Jais 30B

Jais 30B

Jais 30B

G42 & Cerebras

Jais 30B is an open-source large language model developed by G42 and Cerebras, designed to advance Arabic and bilingual NLP research. Trained on over 116 billion Arabic and English tokens, it delivers 83.4% performance on the Arabic MMLU benchmark and supports cross-lingual reasoning, translation, and text generation. Jais 30B leverages a specialized tokenizer optimized for Arabic script, ensuring accurate morphological understanding and natural context flow. With its bilingual training and cultural adaptation, Jais 30B stands as the most powerful Arabic-English model for developers, researchers, and AI startups focusing on regional NLP solutions.

Natural Language Processingai-modelsllm
0
Modelopen source

xLSTM 1.5B

xLSTM 1.5B

xLSTM 1.5B

NX-AI

xLSTM 1.5B is an innovative language model developed by NX-AI that introduces exponential gating mechanisms to extend sequence modeling beyond transformer limits. It outperforms traditional transformer architectures on long-context benchmarks like PG-19 with a perplexity score of 18.2, enabling efficient reasoning across sequences up to 1 million tokens. The model offers 5× faster training, reduced memory consumption, and stable scaling for long document processing, time-series analysis, and contextual learning tasks. With open-source availability and efficient GPU utilization, xLSTM 1.5B is a next-generation alternative for developers and researchers building long-context natural language systems.

Natural Language Processingai-models
0