Factorial experiments with over 1300 runs falsify the hypothesis that INT6 QAT needs a different LR schedule from higher precision and identify a 50M-parameter boundary for INT4 schedule sensitivity.
https://arxiv.org/abs/2510.14823
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Mapping the Schedule x Bit-Width Boundary in Sub-100M Quantisation-Aware Training
Factorial experiments with over 1300 runs falsify the hypothesis that INT6 QAT needs a different LR schedule from higher precision and identify a 50M-parameter boundary for INT4 schedule sensitivity.