List your company

Quantization Distribution Evaluation on C4 (calibration set)

0KL Divergence (Top 10)BF16 (baseline)
-0.0028080.0161460.03510.054054May 12, 2026

Evaluation results

MethodLinks
BF16 (baseline)Precision=BF162026.0501000
PO2(Split87)Quantization=W4A42026.050.036492.350.0464
PO2(NF4)Quantization=W4A42026.050.040391.960.0514
Split87Quantization=W4A42026.050.041391.860.0524
BOF4Quantization=W4A42026.050.044691.550.0568
IF4Quantization=W4A42026.050.045691.420.0585
NF4Quantization=W4A42026.050.04691.410.0589
SFP4Quantization=W4A42026.050.048891.180.0621
4over6Quantization=W4A42026.050.049191.160.0628
NVFP4Quantization=W4A42026.050.054390.670.0696
NVINT4Quantization=W4A42026.050.070289.250.0919