“seem to only manifest in models with extremely large numbers of parameters”
You can do the same with models trained on your laptop in a few seconds. The trick here is to let the model have attention to what is learned, and can also be used on images and other type of data.
The benefit of a lot of parameters is more related to training in parallel, faster and “remember” more from the data.
You can do the same with models trained on your laptop in a few seconds. The trick here is to let the model have attention to what is learned, and can also be used on images and other type of data.
The benefit of a lot of parameters is more related to training in parallel, faster and “remember” more from the data.