implementation detail · filed under model architecture
Indexer Warmup
A warm-up stage used to initialize the index branch and avoid early random selections.
Also called indexer warmup with full attention.
- 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.
We use a two-stage training schedule to initialise the Index Branch and avoid early random selections.
usedtraining objectivein MiniMax-M3MiniMax
a short indexer warmup
usedoptimizationin MiniMax Sparse AttentionMiniMax
Filed alongside
Other methods under model architecture :: token mixer :: sparse attention :: sparse attention indexer.
Compressed Sparse Attention 2Hierarchical Sparse IndexerIndexCacheIndexShareLightning IndexerReindex ModeReuse ModeDense Warm-up StageFine-grained token selectionIndex BranchReuse QSA index selection across speculative decoding stepsAverage poolingBlock-causal scoringCompressed lightweight indexerCross-stage shared-state management for attention reuseDetached indexer-input optimizationIndex Branch outputIndex Branch value headIndexPoolMQA indexerSingle-head index key