DLight comments

Repositories
Issues
Comments

Results 6 comments of


                                            DLight

Best work, which paper I should cited for using this code?

same here @zllrunning

预训练放开vision encoder，效果很差

hi，如果直接全量放开可能会影响本身vit的能力，所以我们使用了layer wise lr decay让vit的浅层改动较小，深层adapt到新的任务上，这部分code在整理中，近期会开源

question about multi-gpu example?

The size of tensor a (1377) must match the size of tensor b (1376) at non-singleton dimension 3 this is commonly caused by the transformer version difference, try transformers==4.33.2

InternLM-XComposer2-4KHD-7B 多卡推理

https://github.com/InternLM/InternLM-XComposer?tab=readme-ov-file#inference-on-multiple-gpus try this one

Meaningless answer when running when running simple examples provided by the authors

it seems weird, is the image path correct?

使用多GPU推理时遇到错误

试一下把transformers版本改成4.33.1