[Question] Why I got nothing when I tested my lora finetune model

[Wenwen Cai]

Question

1.I use finetune_lora to finetune the model for 3 epochs, and I modified the save code to ensure every epoch will save non_lora_trainable.bin
2. I merge lora weight for every epoch
3. I tested on model_vqa but my output tensor is null

[Vignesh-Valaboju]

Are your projector weights changing after every epoch?