How to change GPU

[danboshuiyan]

Your work is very meaningful， which I am very interested in. When I was using it, I found that my GPU0 card did not have enough memory. I want to replace it with 1 card. Where should I modify it?
Thanks for your reply

[blacktanktop]

I honestly don't know much about the GPU replace.
Since JAX is basically being used, I assume it is an issue regarding the use of GPUs in JAX.

The following JAX issue may be helpful (it talks about the use of multiple GPUs)
jax-ml/jax#9859

[danboshuiyan]

I honestly don't know much about the GPU replace. Since JAX is basically being used, I assume it is an issue regarding the use of GPUs in JAX.

The following JAX issue may be helpful (it talks about the use of multiple GPUs) google/jax#9859

Thanks for your reply, I use another computer to make it.
If I want to appoint the residue which the protein created can dock, do I need to modify something to make it?

[blacktanktop]

maybe you can use hotspot option. (but Sorry, I have not checked the behavior. It is no different than the original AfDesign (from about late 2022-02).)

The actual code is as follows, so try it as the JAX code, as in the example, in a format that is acceptable, even if you insert it in the "H" part
https://github.com/ohuelab/Solubility_AfDesign#for-binder-hallucination-can-i-specify-the-site-i-want-to-bind

Solubility_AfDesign/design.py

Lines 223 to 230 in 63de529

	if self._hotspot is None:
	pae_inter = 0.5 * (pae_loss[...,:T,T:].mean() + pae_loss[...,T:,:T].mean())
	con_inter = con_loss[...,:T,T:].mean()
	else:
	# restrict the inter loss to specific hotspot residues
	H = self._hotspot
	pae_inter = 0.5 * (pae_loss[...,H,T:].mean() + pae_loss[...,T:,H].mean())
	con_inter = con_loss[...,H,T:].mean()