File size: 734 Bytes
0393205 |
1 2 3 4 5 6 7 |
# ExLlamaV2 Measurements
Measurement files used to quantize models uploaded here. Depending on the model type one of the following datasets is used for calibration:
* General chat or instruct model: ThePile [0007.parquet](https://huggingface.co/datasets/EleutherAI/the_pile_deduplicated/resolve/refs%2Fconvert%2Fparquet/default/train/0007.parquet)
* Coding model: [WizardLM_evol_instruct_V2_196k](https://huggingface.co/datasets/WizardLM/WizardLM_evol_instruct_V2_196k)
* RP model: pippa.parquet (from users of TheBloke Discord)
* Japanese model: filtered text from file [Japanese Wiki file jawikibooks-20231023-cirrussearch](https://dumps.wikimedia.org/other/cirrussearch/20231023/jawikibooks-20231023-cirrussearch-content.json.gz) |