WFJ

> Can you try to open the image with `PIL.Image.open('your_path')`? Thank you for your reply. There is indeed a problem with the image and it cannot be opened. I found...

>  hi team, I am trying to reproduce mof(clip-vit+dinov2-vit) and Vicuna-7b. And i am facing the error shown in the image. And here is the training settings.  Could...

> 训qwen2-1.5B，保存模型、和评测模型的地方需要改一下，改动地方我们没更新上来，因为只有qwen2有这个问题。所以请微信群里详聊。 您好，我也添加了微信，也一直没通过，烦请通过指导下。谢谢！

This is the loss in the fine-tuning phase. I only changed the batch_size in the hyperparameters. I used four 3090 GPUs. I don't know where the problem lies. deepspeed --include...

> Hi. After pretraining, the initial loss in finetune stage should starts from about 2.5. It seems the problem came from the pretraining stage. Please provide your params in pretrain.sh....

Thank you very much for your reply, per_device_train_batch_size be set 4 when running with a machine of 4 3090GPUs，Otherwise it will oom. I'll try again with a different learning rate....