llmixer's picture
Create README.md
426ccdf verified
metadata
license: llama2
language:
  - en
pipeline_tag: conversational
tags:
  - 4.0bpw
  - h8
  - exl2

Exllamav2 4.0bpw h8 quant for BigWeave-v6-90b.

Calibration dataset: llmixer/20k_random_data

Fits 48GB VRAM with 4k context. Slightly lower ppl than Goliath-120b 3.0bpw.