turboderp's picture
Update README.md
fa3fa06 verified
metadata
license: apache-2.0
base_model: swiss-ai/Apertus-70B-Instruct-2509
base_model_relation: quantized
quantized_by: turboderp
tags:
  - exl3

EXL3 quants of Apertus-70B-Instruct-2509

2.00 bits per weight
2.50 bits per weight
3.00 bits per weight
3.50 bits per weight
4.00 bits per weight
5.00 bits per weight
6.00 bits per weight

Quick sanity checks:

. MMLU 95% CI
2.0 bpw 58.90% +/- 1.50%
2.5 bpw 64.20% +/- 1.46%
3.0 bpw 67.00% +/- 1.43%
3.5 bpw 67.70% +/- 1.43%
4.0 bpw 69.40% +/- 1.40%
5.0 bpw 70.30% +/- 1.39%
6.0 bpw 69.60% +/- 1.40%