general family · filed under optimization
Behavioral regularization
A family of mechanisms introduced to improve RL training stability and encourage efficient, reliable behavior.
- source
- 1
- models
- 2
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
To improve RL training stability and encourage efficient, reliable behavior, we introduce two complementary mechanisms.
usedpost trainingin MiMo-V2.6-Pro and MiMo-V2.6-FlashXiaomi
Filed alongside
Other methods under optimization :: training stability.
Weight paddingActivation or logit clippingAnti-hallucination trainingCross-replica model-weight hash consistency checksElevated constant-learning-rate training stability stress testExponential moving average of checkpointsGradient clippingLearning-rate elevation for training stability stress testingLoss masking for excessively stale tokensOff-policy sample filteringPer-token regularization for off-policy RLSafeguards against training drift and reward hackingSoft droppingTraining stability stress testingWeight clippingWeight decay coupled to learning-rate squared