train a GPTQ quantised LLM with a local or hosted custon dataset using PEFT (parameter efficient fine tuning) and/or run inference with/without the model adapter, this useful for multiple fine-tuned agents. The generated adapter is here sujitvasanth/TheBloke-openchat-3.5-0106-GPTQ-PEFTadapterJsonSear The custom training dataset used is: https://huggingface.co/datasets/sujitvasanth/jsonsearch2 there is a .csv version in this repository
run GPTQ-finetune.py adjusting the location of your GPTQ model, csv file and where to save the generated model adapter the example dataset improves the json search capabilities of The Bloke's openchat GPTQ
