I want to retest the code using a model with fewer parameters, but I’ve noticed that the --num-iters setting affects the final perplexity. I currently believe that --num-iters is a parameter that controls the amount of data to be judged, but I’m not certain. I want to know how I should set this parameter. The project provides dense model testing and sparse model testing for models with 66b and 175b parameters. I am a bit confused about the setting of --num-iters, and I hope to get your help. How should I set this parameter for models like 1.3b and 6.7b?
I want to retest the code using a model with fewer parameters, but I’ve noticed that the --num-iters setting affects the final perplexity. I currently believe that --num-iters is a parameter that controls the amount of data to be judged, but I’m not certain. I want to know how I should set this parameter. The project provides dense model testing and sparse model testing for models with 66b and 175b parameters. I am a bit confused about the setting of --num-iters, and I hope to get your help. How should I set this parameter for models like 1.3b and 6.7b?