Jingyao Li

Hi, our reported *pass@1* is the average/normalized *pass@1*. You can refer to [this benchmark paper](https://arxiv.org/abs/2105.09938) for the detailed metric definition.

Thank you for bringing this to our attention. We have verified that what you mentioned is correct. Due to an error in reporting in line with previous work, we have...

Thank you for bringing this to our attention. We have aligned the metrics and update the tables.

Sure, we have uploaded our motcode dataset [here](https://huggingface.co/datasets/JingyaoLi/MoTCode-Data). Hope it's helpful

This is due to a mismatch in shapes between your `image_train.pkl` and the `Image_list`. This may be because your ImageNet dataset is incomplete, resulting in a dimension of 91710 instead...

Hi, we have released all the pre-trained weights and [MOODv2](https://github.com/dvlab-research/MOOD/tree/master/MOODv2). You can find the MoCov3 pre-trained on ImageNet-22k [here](https://github.com/dvlab-research/MOOD/tree/master/MOODv2#pretrained-model-preparation).

Hi! The reason is that there is an overlap between ImageNet-30 and ImageNet-1k. Consequently, if we conduct fine-tuning on ImageNet-30, it would transform into a straightforward binary-classification task rather than...

Hi, [here](https://openaccess.thecvf.com/content/CVPR2023/supplemental/Li_Rethinking_Out-of-Distribution_OOD_CVPR_2023_supplemental.pdf) is the link to the appendix.

hi, you can use our models in `qllm/models` like `from qllm.models import LlamaForCausalLM` (This will allow the param `question_ids` in `forward`) The complete code is as following: ``` import torch...

我们定义了GreedySearch类，您可以定义searcher=GreedySearch(model, tokenizer)并用searcher.generate(question_ids=xxx)替代model.generate()