implementation detail · filed under model architecture
Modality-specific auxiliary-loss-free load balancing
Extends auxiliary-loss-free balancing by maintaining separate expert correction biases for different modalities, including text and image tokens.
Also called modality-specific load balancing.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- core
How sources treat it
One count per evidence span, weakest treatment to strongest.
core 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
we extend auxiliary-loss-free load balancing (Wang et al., 2024a) by maintaining separate expert-wise correction biases for text and image tokens.
coreoptimizationin DeepSeek-V4.1-FlashDeepSeek
Filed alongside
Other methods under model architecture :: channel mixer :: mixture of experts :: expert load balancing.
Auxiliary-loss-free load balancingQuantile BalancingHistogram-based quantile estimationAuxiliary-loss load balancingExact coordinate minimizationExpert bias update factorExpert Parallelism Load BalancingExponential moving average of estimated quantilesformHCLoad balancingMaxVioPersistent load balancingPooled-global-batch quantile estimationRound-robin load balancing