dream.py issue on M1 Mac: torch._C' has no attribute '_cuda_resetPeakMemoryStats'

[yudhanjaya]

M1 Mac. After model load,

Loading model from models/ldm/stable-diffusion-v1/model.ckpt
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
making attention of type 'vanilla' with 512 in_channels
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
making attention of type 'vanilla' with 512 in_channels
Using half precision math. Call with --full_precision to use more accurate but VRAM-intensive full precision.
>> Setting Sampler to k_lms
model loaded in 7.59s
* Initialization done! Awaiting your command (-h for help, 'q' to quit)
dream> "a tower of books"

this error happens:

Traceback (most recent call last):
  File "/Users/yudhanjaya/stable-diffusion/scripts/dream.py", line 518, in <module>
    main()
  File "/Users/yudhanjaya/stable-diffusion/scripts/dream.py", line 99, in main
    main_loop(t2i, opt.outdir, opt.prompt_as_dir, cmd_parser, infile)
  File "/Users/yudhanjaya/stable-diffusion/scripts/dream.py", line 208, in main_loop
    image_list  = t2i.prompt2image(image_callback=callback, **vars(opt))
  File "/Users/yudhanjaya/stable-diffusion/ldm/simplet2i.py", line 283, in prompt2image
    torch.cuda.torch.cuda.reset_peak_memory_stats()
  File "/Users/yudhanjaya/miniconda3/envs/ldm/lib/python3.10/site-packages/torch/cuda/memory.py", line 260, in reset_peak_memory_stats
    return torch._C._cuda_resetPeakMemoryStats(device)
AttributeError: module 'torch._C' has no attribute '_cuda_resetPeakMemoryStats'

I've done all the steps in trouble-shooting: as far as I can tell, loading is not an issue. What am I doing wrong?

[sergiocasero]

I can reproduce it on MAC M1 pro, if I just run img2text directly works as expected

[rcknight]

Same issue here - txt2img script works fine but dream fails with this error

[junukwon7]

Seems like this is a kind of a patch, however returning black squares.

[personjerry]

Just comment out torch.cuda.torch.cuda.reset_peak_memory_stats() on line 283 of simplet2i.py

# torch.cuda.torch.cuda.reset_peak_memory_stats()

[yudhanjaya]

Can confirm that the original scripts work: I just generated Sheldon Cooper as a GTA character, no issues with txt2img.py.

Just comment out torch.cuda.torch.cuda.reset_peak_memory_stats() on line 283 of simplet2i.py

# torch.cuda.torch.cuda.reset_peak_memory_stats()

Thanks! Just tried this, but I don't seem to be getting output. It still seems to be trying to use cuda - but the 'it' here appears to be dream.py (as best as I can make out) and not the sd scripts themselves.

dream> "Cat Pharaoh dreaming of pyramids"
Generating:   0%|                                                                   | 0/1 [00:00<?, ?it/s]
"LayerNormKernelImpl" not implemented for 'Half'
Are you sure your system has an adequate NVIDIA GPU?
Usage stats:
   0 image(s) generated in 0.01s
   Max VRAM used for this generation: 0.00G
   Max VRAM used since script start:  0.00G

[lstein]

This is fixed in pull request #256. Just waiting for a code review to merge.

[yudhanjaya]

This is fixed in pull request #256. Just waiting for a code review to merge.

Thanks, much appreciated!