oa::VramBudgetConfig

VRAM BUDGET — auto-tune batch_size * seq_len from free VRAM (implementation in budget.cpp)

Public Data Members

bytesPerParamoa::I32field
maxBatchSizeoa::I32field
maxSeqLenoa::I32field
minBatchSizeoa::I32field
minSeqLenoa::I32field
optimizerStatesPerParamoa::I32field
preferredBatchSizeoa::I32field
preferredSeqLenoa::I32field
safetyMarginPercentoa::F32field