| datasets: | |
| - ehartford/dolphin | |
| - jondurbin/airoboros-2.2.1 | |
| - ehartford/samantha-data | |
| - ehartford/WizardLM_evol_instruct_V2_196k_unfiltered_merged_split | |
| language: | |
| - en | |
| license: other | |
| license_name: yi-license | |
| license_link: LICENSE | |
| quant of [ehartford's](https://huggingface.co/ehartford) [dolphin-2.2-yi-34b-200k](https://huggingface.co/ehartford/dolphin-2.2-yi-34b-200k) | |
| done with exllamav2 0.0.11 so should have lower loss than earlier quants | |
| ``` | |
| python3 convert.py \ | |
| -i /input/ehartford_dolphin-2.2-yi-34b-200k/ \ | |
| -c /input/pippa_cleaned/0000.parquet \ | |
| -o /output/temp/ \ | |
| -cf /output/dolphin-2.2-yi-34b-200k-4bpw-h6-exl2/ \ | |
| -l 8192 \ | |
| -ml 8192 \ | |
| -b 4 \ | |
| -hb 6 | |
| ``` |