turboderp's picture
Update README.md
0e09615 verified
metadata
license: mit
base_model: z-lab/Qwen3.6-27B-DFlash
base_model_relation: quantized
quantized_by: turboderp
tags:
  - exl3

EXL3 quants of Qwen3.6-27B-DFlash

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

Quant Mean acc. tokens¹
2.50 bpw 4.04
3.00 bpw 4.34
3.50 bpw 4.22
4.00 bpw 4.46
5.00 bpw 4.36
6.00 bpw 4.43
BF16 4.12

¹ Mean verified tokens per 15-token draft, CatBench at temp=0, using 4.15bpw target model