Lianghui Zhu

> Thank you for your interest in our work. > > For your first question, DeepLab Large-FOV is equal to DeepLab-V1, and DeepLab-ASPP is equal to DeepLab-V2. > The DeepLab-V2...

When I use the scripts like below: ``` torchrun --nnodes=1 --nproc_per_node=1 --master_port=25301 \ llava/train/train_mem.py \ --model_name_or_path /home/zhulianghui/ProjectC_ChatGPT/alpaca-quan/output/vicuna-13b-composite-3gpt-data \ --data_path /home/zhulianghui/ProjectC_ChatGPT/llava/reference/LLaVA-CC3M-Pretrain-595K/chat_new.json \ --image_folder /share/project/lianghuizhu/cc3m-llava \ --vision_tower /home/zhulianghui/ProjectC_ChatGPT/llava/reference/clip-vit-large-patch14 \ --tune_mm_mlp_adapter True \...

> Hi @Unrealluver Are you still facing this issue? Thanks, I have found the reason.

> Hi @Unrealluver, the issue has been fixed, and I trained for very few iterations (400 iterations) with LLaVA-7B checkpoint, and verified the results are reasonable. > > We have...

Thanks for your reply.

Hi, I am here to clarify the reason for the abnormal response. I noticed that in your latest code, we could get the true response, you switch the transformer library...

@haotian-liu Thanks, I am waiting for your further reply.

Hi Jiarun, For the first question, you can regard `fused_add_norm` as a fast version of your displayed code. We also use the `LN -> Mixer -> Add` for the second...

Greetings! Thanks for your suggestion! We have fixed it.

We are preparing the code. It will be released soon~