implementation detail · filed under training objective
Gradient detachment
Stops gradient propagation at the Index Branch input.
Also called Gradient Detach, gradient detach for KL loss.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 2
Documented in
Evidence
2 spans quoted from the sources, strongest treatment first.
apply stop-gradient to the Index Branch input
usedtraining objectivein MiniMax-M3MiniMax
stopping the KL gradient at the Index Branch input
usedoptimizationin MiniMax Sparse AttentionMiniMax
Filed alongside
Other methods under training objective :: auxiliary loss.