Redefining residual errors to include compensation-aware discrepancies and realigning calibration to full-precision outputs improves GPTQ and GPTAQ performance on LLMs.
Subsequently, we analyzed the influence of the calibration set Table 12: Performance of 3-bit per-group symmetric weight-only quantization
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
-
Rethinking Residual Errors in Compensation-based LLM Quantization
Redefining residual errors to include compensation-aware discrepancies and realigning calibration to full-precision outputs improves GPTQ and GPTAQ performance on LLMs.