Calibration dataset used to perform GPTQ

#23
by ht-rohit - opened

Hi, what datasets have been used in order to calibrate the quantization of a Falcon-40b-instruct model (4-bit GPTQ or any bits version, or any other model)?

How does this data affect the performance (accuracy) of the model on inference or further instruction-tuning tasks for a particular domain?

Sign up or log in to comment