Shiyu

I'm using 8 V100 to reproduce yolox-s coco training and find the training time is much slower than your shared log [here](https://github.com/Megvii-BaseDetection/YOLOX/issues/735#issuecomment-940607639). And I find that `--fp16` doesn't speed up...

Add Segment Anything Model (SAM) in MLX.

ADE B3 model SS mIoU doesn't match paper Table 1.

3

Run `python tools/test.py local_configs/segformer/B3/segformer.b3.512x512.ade.160k.py segformer.b3.512x512.ade.160k.pth` with given pretrained checkpoint in this repo. And got ``` +--------+-------+-------+-------+ | Scope | mIoU | mAcc | aAcc | +--------+-------+-------+-------+ | global | 45.48...

Hi, it seems `window_block_indexes` is set as a nested list when building the model https://github.com/baaivision/Painter/blob/07c8a2a04d8fab791dae804cd7509a3e1cdaa321/Painter/models_painter.py#L481 in which case `window_size` in this line is always 0 https://github.com/baaivision/Painter/blob/07c8a2a04d8fab791dae804cd7509a3e1cdaa321/Painter/models_painter.py#L307 Is this expected?

Large memory occupation

7

Hi, I'm training faster-rcnn on 4 gpus with coco dataset converted to LMDB. I used num_worker=4 for the dataloader and I found that the memory occupation is almost 60Gb. I...

Unmentioned but critical LayerNorm

5

To achieve comparable result as the original paper. LayerNorm is applied to the feature before NF. This is never mentioned in the paper and the usage is very tricky (but...

Unmatched model A.d. parameter

12

The model additional parameter number cannot match Table 1. in the paper. | | wide-resnet-50 | resnet18 | DeiT | CaiT | | ----------------- | -------------- | --------------- | ---------------...

support 3D attention mask in bert

8

# What does this PR do? Fixes #31036 Try to fix with minimal change as mentioned in #31302 3D attention masks are used in [GroundingDINO](https://github.com/IDEA-Research/GroundingDINO) ## Before submitting - [...