Unverified70% confidenceFactExact time
Mixed-precision training was jointly proposed by NVIDIA and Baidu in 2018; its core idea is to use FP16 in forward and backward propagation to reduce VRAM usage and accelerate computation, while retaining an FP32 master weight copy for parameter updates.
2
Sources
70%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
JAX主机卸载技术：突破LLM训练显存瓶颈的实战指南
Related Claims
Unverified混合精度训练由 NVIDIA 在 2018 年提出，同时使用 FP32 和 FP16 两种精度81% similarVerifiedIn mixed-precision training, the forward pass uses FP16 to leverage Tensor Core acceleration while parameter updates retain FP32 to maintain numerical stability, which has become standard practice for large model training.78% similarUnverifiedAMP（自动混合精度训练）通过将部分计算从 FP32 降至 FP16 可将 GPU 有效计算吞吐量提升 2-3 倍并减少显存占用75% similarUnverifiedLTX-Video 引入高效 VAE 压缩和流匹配（Flow Matching）训练目标，使 22B 参数模型在消费级 GPU 上实现近实时推理73% similarUnverified开发者采取混合精度策略，将对质量敏感的层保留在BF16，仅对能从FP8获益的部分做转换72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/503799API
curl https://kongchang.com/api/v1/knowledge/claims/503799MCP
get_claim(id=503799)