GPTQ quantisation of ai-forever/ruGPT-3.5-13B · Hugging Face
Small perplexity test: before quantization - ‘mean_perplexity’: 10.241 after quantization - ‘mean_perplexity’: 10.379
Data - RussianSuperGlue > DaNetQA/train.jsonl[‘passage’]
As this is a hastily thrown together quant with no prior experience in quants, use TheBloke (Tom Jobbins) version if he releases a quant for this model.