Benjamin Bossan
Benjamin Bossan
So what you're saying is that the slowness stems from the lackluster support of MPS in PyTorch, and as PEFT is using PyTorch, slow MPS performance is expected. Is that...
All right. We can keep this open for the time being, maybe it helps get some eyes on the topic.
Could you please give more context, what are you referring to exactly and where is this new parameter being used? Also, as is, this PR assumes that the base layer...
Ah I see, thanks for the pointers. So this was added to megatron more than a year ago, so I guess it should be fine, but I'm not sure if...
@dhrhank187 could you please merge with/rebase on `main` branch? That should fix the failing CI.
@dhrhank187 For some reason, the latest change resulted in a huge diff with 56 files being touched. I think if you rebase on main, this should be fixed. Could you...
@dhrhank187 do you still plan on working on this?
Thanks for the offer @ParagEkbote. I think your suggestion should be enough.
Thanks for reporting this bug and providing a potential solution. Would you be interested in creating a PR with your fix?
This issue is resolved via #2104.