#training frameworks
7 stories · 0 in 7 days · 1 in 30 days
Stories, products and papers tagged "training frameworks", newest first.
Oct 30
Oct 4
AI2 Open-Sources Olmo-core Training Library for OLMo Models
Teams that want full control over pretraining can now start from AI2's scripts instead of assembling their own distributed stack, which cuts weeks of infrastructure work. It matters for open-source researchers comparing OLMo against Llama or Qwen, since OLMo's data and training code are fully reproducible. Practitioners can clone the repo and run the provided annealing scripts on their own data.
Jul 14
Chinese contentPrime Intellect 训练栈:Verifiers、PrimeRL 与后训练算法详解
想了解 Prime Intellect 的异步 RL 框架和模块化验证系统吗?看看他们如何让长程智能体训练更高效。
Jun 24
Models12:00
Chinese contentBluTrain:基于C++/CUDA的AI训练框架这个新框架用C++从头写,训练GPT-2比PyTorch快3%且省内存22%,适合追求极致性能的开发者。
May 27
Chinese content面壁智能开源全球首个AI编写生产级训练框架ForgeTrain
ForgeTrain证明了AI可以自主编写生产级训练框架,解决了大模型训练对人工编码的依赖问题。做模型训练或国产芯片适配的团队可以直接用这套开源工具链复现顶尖模型,建议点开看看具体实现。
IT之家Source
May 26
Chinese contentModel Best 开源 BitCPM-CANN:国产芯片实现 1.58-bit 训练
国产芯片终于能跑 1.58-bit 训练了,内存需求直降 6 倍,用国产算力的 AI 团队可以直接上手试试。
Chinese contentBingbi AI 开源 BitCPM-CANN:1.58-bit 训练框架,国产算力可用
国产算力生态终于有了低精度训练的利器——1.58-bit 训练框架让国产芯片也能高效跑大模型,做国产 AI 部署的团队可以直接拿来用,内存省六倍,值得关注。