Spikes in co-training #92
Description
Hi! Just wondering whether you also experienced the spikes in training loss curve when lr is 1e-6 for qwen2-vl-2b?
I saw in train.md your lr is 1e-4, have not tried it yet.
{{ message }}
Hi! Just wondering whether you also experienced the spikes in training loss curve when lr is 1e-6 for qwen2-vl-2b?
I saw in train.md your lr is 1e-4, have not tried it yet.