implementation detail · filed under optimization
Off-policy sample filtering
Excludes samples that fail due to environment collapse, with failure reasons recorded.
Also called Dropping off-policy and noisy samples.
- source
- 1
- model
- 1
- 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.
we record the failure reason for each sample and exclude samples that fail due to environment collapse
usedoptimizationin GLM-5Z.ai
Filed alongside
Other methods under optimization :: training stability.
Weight paddingActivation or logit clippingAnti-hallucination trainingBehavioral regularizationCross-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 tokensPer-token regularization for off-policy RLSafeguards against training drift and reward hackingSoft droppingTraining stability stress testingWeight clippingWeight decay coupled to learning-rate squared