Segmentation Fault in Mamba
bsxxdw opened this issue · comments
Hello authors!
I've observed that you've attempted to train Mamba on image classification tasks. I'm curious: have you encountered the segmentation fault
error during this process?
I've been training Mamba on ImageNet1K, but it appears to be quite unstable. After several epochs, I encounter a segmentation fault
that interrupts the training process. Have you experienced this issue as well?
Hi @bsxxdw - nope I did encounter any errors like that - I trained cifar-10 on maybe 100 epochs and everything ran smoothly.
@apapiu Thanks for your reply! BTW, would you mind to share some information about your training infrastructure (e.g., GPU cards, number s of CPU cores and system memory), this will help me a lot.
Great thanks in advance!
I was training it on a single T4 GPU on Colab with the high memory flag turned on (I think that's 8 cores). Hope that helps!