algo
now
.net
new ·
18
pairs
atlas
problems
fields
listen
quant
AI
philosophy
algonow
/
algorithms
/
Mixed-precision training
Mixed-precision training
Pairings in the atlas
Loss scaling
standard
Low-precision training
llm-training-alignment
Rivals: other methods for the same problems
FP8 training
Fully sharded data parallel
Gradient accumulation
Gradient checkpointing
ZeRO
Where it sits
llm-training-alignment
·
Machine Learning & AI