Relation between model parameters and batchsize

1

I am trying to find relation between batch size and total parameter?

  • Can i increase batchsize of inputs in inference by reducing the model parameters?,
    i tried but could not see any difference :thinking:.

Reducing parameters may refer to prunning of model, consider this topic as discussion

2

What exactly are you trying to achieve? The only relation between batchsize and number of parameters is the memory they occupy (for a fixed amount of GPU memory if your model is larger, the maximum batchsize will be lower)

3

i am just trying to figure out relation between these two parameters.
So for example:
model has parameters 100k which has batch size of 16
then optimized model has 80k can i feed batch size greater than 16 here?? is it correct

4

Yes Iā€™d think so. If you have less parameters in your memory, you can pass a larger batch size.

1 Like