GRPO News - Blockchain.News

DEEPSEEK

DeepSeek is an AI company and a family of large language models based in Hangzhou, China. It was founded in 2023 and funded by High-Flyer, a well - known quantitative asset management giant. DeepSeek is dedicated to developing advanced large language models and related technologies. It has released several models, including DeepSeek LLM, DeepSeek Coder, DeepSeekMath, and DeepSeek - VL. The latest version, DeepSeek - V3, which was launched in December 2024, has 67.1 billion parameters and was trained on a dataset of 14.8 trillion tokens. It uses FP8 training and open - sources the native FP8 weights. Benchmark tests show that it outperforms Llama 3.1 and Qwen 2.5 while matching GPT - 4O and Claude 3.5 Sonnet. In addition, DeepSeek - R1, which was officially released on January 20, 2025, performs on a par with OpenAI O1 in terms of mathematics, code, and natural language reasoning tasks. DeepSeek's models have a wide range of applications, such as chat and coding scenarios, multilingual automatic translation, image generation, and AI painting. With their high performance and low cost, DeepSeek's models have quickly gained popularity. For example, on February 2, 2025, the DeepSeek app climbed to the top of the download charts in 140 countries on the Apple App Store and also topped the Android Play Store in the United States

deepseek

NVIDIA NeMo-RL Utilizes GRPO for Advanced Reinforcement Learning

NVIDIA introduces NeMo-RL, an open-source library for reinforcement learning, enabling scalable training with GRPO and integration with Hugging Face models.

by Peter Zhang
Jul 10, 2025

DEEPSEEK

NVIDIA NeMo-RL Utilizes GRPO for Advanced Reinforcement Learning

Trending topics