ComfyUI KSamplerAdvanced Tensor Size Mismatch and Out of Memory Errors

[boredcook77]

Hello everyone,

I've been working with ComfyUI and have encountered a series of errors that I'm unable to resolve. The main issues are as follows:

I receive a "tensor size mismatch" error stating, "The size of tensor a (8) must match the size of tensor b (0) at non-singleton dimension 0." Subsequently, I encounter an "out of memory" error with details indicating that the allocation on device 0 would exceed allowed memory. These errors persist despite attempting various fixes, such as adjusting batch sizes and ensuring that all nodes and dependencies are up to date. It appears that somewhere along the line, the tensors being processed are not aligning correctly, and my system is running out of allocatable memory to execute the tasks.

I'm reaching out to see if anyone has insights into which specific nodes or settings within ComfyUI could be tweaked to address these issues, or any tips on memory management for large tensor operations within this context.

Any guidance or suggestions would be greatly appreciated!

[jokero3answer]

I'm also experiencing this error, it's been a couple of days now... please ask the author to fix it as soon as possible...

[jokero3answer]

@tobbez @jeske @realazthat

[tzhh11]

so am i

[tzhh11]

I've encountered this problem before. It took me a long time to solve it. First, you need to look at the number of pictures in the workflow on the path processed by your clip. For me it's controlnet processing. Look at your split image. The image group you are processing is image a or image b. Assume that the total number of pictures you have is 1 and your split index is 10. Then all pictures will be in image a, and image The number of pictures in group b is 0, and if you select group b when processing images, this problem will occur. That is, the so-called tensor b is 0.