Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

KV Cache Quantization on WikiText-2 prompt 2048-token (train)

0.0026Softmax KL Divergence

Block-GTQ

-0.0759960.4545270.985051.515573Jun 23, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2026.06
0.0026
2026.06
0.0035
2026.06
0.0041
2026.06
0.0056
2026.06
0.0063
2026.06
0.0088
2026.06
0.0088
2026.06
0.0091
2026.06
0.0098
2026.06
0.0099
2026.06
0.0104
2026.06
0.0121
2026.06
0.014
2026.06
0.0151
2026.06
0.0153
2026.06
0.0166
2026.06
0.0214
2026.06
0.0245
2026.06
0.0271
2026.06
0.0327
2026.06
0.0332
2026.06
0.0335
2026.06
0.0367
2026.06
0.0378
2026.06
0.0405
2026.06
0.0435
2026.06
0.0444
2026.06
0.049
2026.06
0.051
2026.06
0.0515
2026.06
0.054
2026.06
0.0546
2026.06
0.0548
2026.06
0.0558
2026.06
0.0567
2026.06
0.0569
2026.06
0.0571
2026.06
0.0584
2026.06
0.0594
2026.06
0.0614
2026.06
0.0663
2026.06
0.068
2026.06
0.0684
2026.06
0.0696
2026.06
0.0769
2026.06
0.0773
2026.06
0.0829
2026.06
0.0859
2026.06
0.0989
2026.06
0.1024
2026.06
0.1072
2026.06
0.1188
2026.06
0.121
2026.06
0.1229
2026.06
0.1234
2026.06
0.1256
2026.06
0.1274
2026.06
0.1288
2026.06
0.1298
2026.06
0.1444
2026.06
0.1596
2026.06
0.1617
2026.06
0.1636
2026.06
0.178
2026.06
0.1799
2026.06
0.196
2026.06
0.202
2026.06
0.2023
2026.06
0.2097
2026.06
0.2118
2026.06
0.2288
2026.06
0.2544
2026.06
0.2915
2026.06
0.3053
2026.06
0.3114
2026.06
0.3275
2026.06
0.3338
2026.06
0.345
2026.06
0.3492
2026.06
0.3594
2026.06
0.4381
2026.06
0.4546
2026.06
0.4708
2026.06
0.484
2026.06
0.5126
2026.06
0.5326
2026.06
0.5569
2026.06
0.5709
2026.06
0.5847
2026.06
0.6349
2026.06
0.6359
2026.06
0.7481
2026.06
0.8097
2026.06
0.8782
2026.06
0.9397
2026.06
1.3615
2026.06
1.7932
2026.06
1.8197
2026.06
1.9675