skt_A.X-3.1-EXL3 / README.md
ArtusDev's picture
Update README.md
b5fa480 verified
metadata
base_model: skt/A.X-3.1
base_model_relation: quantized
quantized_by: ArtusDev
license: apache-2.0
license_link: https://huggingface.co/skt/A.X-3.1/blob/main/LICENSE
language:
  - en
  - ko
model_id: skt/A.X-3.1
developers: SKT AI Model Lab
model-index:
  - name: A.X-3.1
    results:
      - task:
          type: generate_until
          name: mmlu
        dataset:
          name: mmlu (chat CoT)
          type: hails/mmlu_no_train
        metrics:
          - type: exact_match
            value: 75.1
            name: exact_match
      - task:
          type: generate_until
          name: kmmlu
        dataset:
          name: kmmlu (chat CoT)
          type: HAERAE-HUB/KMMLU
        metrics:
          - type: exact_match
            value: 69.2
            name: exact_match
tags:
  - exl3

EXL3 Quants of skt/A.X-3.1

EXL3 quants of skt/A.X-3.1 using exllamav3 for quantization.

Quants

Quant(Revision) Bits per Weight Head Bits
2.5_H6 2.5 6
3.0_H6 3.0 6
3.5_H6 3.5 6
4.0_H6 4.0 6
4.5_H6 4.5 6
5.0_H6 5.0 6
6.0_H6 6.0 6
8.0_H8 8.0 8

Downloading quants with huggingface-cli

Click to view download instructions

Install hugginface-cli:

pip install -U "huggingface_hub[cli]"

Download quant by targeting the specific quant revision (branch):

huggingface-cli download ArtusDev/skt_A.X-3.1-EXL3 --revision "5.0bpw_H6" --local-dir ./