【新智元导读】RedStone是一个高效构建大规模指定领域数据的处理管道，通过优化数据处理流程，从Common Crawl中提取了RedStone-Web、RedStone-Code、RedStone-Math和RedStone-QA等数据集，在多项任务中超越了现有开源数据集，显著提升了模型性能。 过去几年，大型语言模型 ...

The best way to understand deep learning is learning by doing. This open-source book represents our attempt to make deep learning approachable, teaching you the concepts, the context, and the code.

ProgCo enables AI to learn this approach as well. Specifically, AI generates a "verification program" in response to received ...

尽管当前语言模型在语言能力表现出色，但其解决数学问题的能力在现实应用中仍然面临挑战。虽然研究者开发了许多策略和数据集以增强LLMs的数学能力，但在部署的LLM系统中同时保持和提高语言和数学能力仍然是一个挑战。在这项工作中，我们定制了自我批评 ...

Mathstral, a 7-billion-parameter model, was developed by the French startup in collaboration with Project Numina, a non-profit organization focused on advancing human and artificial intelligence in ...