A calibration-free rounding scheme uses a diffusion model trained on a model's own weights to resolve ambiguous midpoint rounding, improving 3- and 4-bit weight-only quantization on small LLMs.
NeurIPS , pages =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
ReRound: Reconstructive Rounding to Resolve Midpoint Ambiguity in Calibration-Free LLM Quantization
A calibration-free rounding scheme uses a diffusion model trained on a model's own weights to resolve ambiguous midpoint rounding, improving 3- and 4-bit weight-only quantization on small LLMs.