FAMO

FAMO copied to clipboard

Whether setting "--scale-y=True" will affect the performance of other methods like MGDA, IMTL?

1

Hi, as I noticed in your previous issue, you said "--scale-y=True" would affect FAMO's performance. Will it affect the other methods' performance, like MGDA, LS, or IMTL in your code?

MarcuXu

Question on FAMO loss behavior

2

Hello and thanks for the fantastic work on FAMO! I've been experimenting with FAMO in some personal projects and have observed an interesting behavior related to loss values. It appears...

MendezEd

Hello, In the toy example [here](https://github.com/Cranial-XIX/FAMO/blob/0a6ad06dbdd55fce73ec01e3a175c2498cd1e5e9/experiments/toy/trainer.py#L59) you have set `self.min_losses = torch.Tensor([-21 * scale, -21])` where `scale = 1e-1` by default. If I understand correctly, this is meant to replicate...

imanfs

Hi, I am using FAMO for a MTRL task. I have a question about the critic loss used in famo.py, line 375: ```real_backprop_loss = torch.sum(weights * (task_loss + 1e-8).log() *...

RomainLITUD

MTRL success rate metrics

8

Hi, I'm trying to replicate the results of MT10 in the paper, I'm just wondering is the final results obtained from the `Su` of the 2M steps eval log? Current...

KlZOnmyway

Hi, I tried the default function from pandas to calculate the mean rank score, however, there is some discrepancy in the results. Could you please provide the way of calculating...

KlZOnmyway