大模型微调速度提升 2 倍、显存减少 70%，单卡也能训练。
Transformer Engine
A library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit and 4-bit floating point (FP8 and FP4) precision on Hopper, Ada and Blackwell GPUs, to provide better performance with lower memory util
- 仓库
- NVIDIA/TransformerEngine
- Stars
- ★ 3,575
- Forks
- 860
- 语言
- Python
- 许可证
- Apache-2.0
- 最近更新
- 2026-10-12
- 访问次数
- 0
PyTorch
同类项目
统一高效微调 100+ 大模型，零代码 Web 界面，国内最流行的微调工具。
统一的 AI 计算引擎，用于分布式训练、调参、推理与服务。
机器学习与 GenAI 生命周期管理平台，含追踪与评测。
🧠 Train a 64M-parameter LLM from scratch in just 2h!
DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.