训练出现除0错误

[gqf123]

用的官方的数据试了一下，一训练就出现除0错误，也不说是哪出错了，该咋办
| Name | Type | Params

0 | transformer | FluxTransformer2DModel | 12.8 B

927 M Trainable params
11.9 B Non-trainable params
12.8 B Total params
51,318.772Total estimated model params size (MB)
Building buckets using %d workers... 1
Building buckets using %d workers... 1
[rank: 0] Received SIGTERM: 15
Bucket Info:
Bucket [#sample, #batch] by aspect ratio:
(%s): #sample: %s, #batch: %s 1:1 561 70
===== Image Info =====
Image Bucket by HxWxT:
%s: #sample: %s, #batch: %s ('512px', 1) 561 70

#image sample: %s, #image batch: %s 561 70
===== Video Info =====
Video Bucket by HxWxT:

#video sample: %s, #video batch: %s 0 0
===== Summary =====
#non-empty buckets: %s 1
Img/Vid sample ratio: %.2f 0
Img/Vid batch ratio: %.2f 0
vid batch 256: %s, vid batch 768: %s 0 0
Vid batch ratio (256px/768px): %.2f 0
#training sample: %s, #training batch: %s 561 70
[ce3fmv77d6ue3-0:3476971:0:3476971] Caught signal 8 (Floating point exception: integer divide by zero)
[ce3fmv77d6ue3-0:3476970:0:3476970] Caught signal 8 (Floating point exception: integer divide by zero)

[bLUe1RuI]

@gqf123
看着像数据集加载的问题，你可以检查下 code，每一列中，第一个为 json 文件，是否已经用新的 json 替换，第二个为 image_root, 可以看看路径与 json 里的文件名，是否组成正确的路径。实在不行的话，可能需要调试看看，有调试问题可继续提~