A soft-thresholding rule equivalent to selecting the top alpha percent of weights by magnitude is presented as a statistical pre-calibration for LLM quantization, with results close to but not clearly better than calibration-based baselines.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Rethinking Post-Training Quantization: Introducing a Statistical Pre-Calibration Approach
A soft-thresholding rule equivalent to selecting the top alpha percent of weights by magnitude is presented as a statistical pre-calibration for LLM quantization, with results close to but not clearly better than calibration-based baselines.