inference/research
verifiedhigh · 0.90updated 2026-10-03

ExLlamaV2 README (EXL2 format description)

repo

Fields

id
sources/quant-exl2-readme
URL
https://github.com/turboderp-org/exllamav2
Publisher
turboderp
Kind
repo
Published
2026-10-03
Accessed
2026-10-03
notes
EXL2 = GPTQ optimization with mixed bit-widths to hit any average bpw between 2 and 8. Repo archived; work continues in ExLlamaV3.

Notes

EXL2 = GPTQ optimization with mixed bit-widths to hit any average bpw between 2 and 8. Repo archived; work continues in ExLlamaV3.

Cross-references

This record references no other records.