Report
A video on training AI models too big for one GPU
The poster says it covers how real numbers are represented and stored, mixed-precision training, ZeRO and FSDP.
TLDR
A post shares a video described as explaining how to train models that cannot fit on one GPU. The poster says it covers how real numbers are represented and stored, mixed-precision training, the Zero Redundancy Optimizer (ZeRO) and FSDP.
Combined views
17
1 Source, first seen ago
