Aitrepreneur patreon

1-Click INSTALL Ai-TOOLKIT WAN LORA TRAINING!

🕑 Added 2025-07-20 16:50:43 +0000 UTC

Comments

nurtu

CUDA error: misaligned address CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

Matthew Smallwood

I'm trying to use this on runpod with an RTX pro 6000. i get this error. "Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!"

totally_not_a_weirdo

"[ERROR] curl not found ΓÇô install Windows 10 v1803+ or add it to PATH." This pops up immediately after choosing which RTX-series card (on the installer). I definitely have curl.exe on my computer, and it sure seems to be added to PATH. ETA> After over 2 years of obsessing over this stuff, today I learned that I have been doing the PATH thing wrong. Having corrected that (those, actually), it installed as expected.

Kent Umeki

will i have better results if i train the model using videos? or does it not matter?

D Dunn

getting this error on runpod CUDA out of memory. Tried to allocate 68.00 MiB. GPU 0 has a total capacity of 31.37 GiB of which 58.69 MiB is free. Including non-PyTorch memory, this process has 31.30 GiB memory in use. Of the allocated memory 30.57 GiB is allocated by PyTorch, and 144.35 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

SAMIRA

I want to train NSFW but for the descriptions, GPT chat doesn't give them to me.

Aitrepreneur

this error actually happens because you don't have enough space on the pod

Patrick Brown

I did this last night and trained a Lora for Lake Bell and it turned out amazing! Thank you! However, today, I'm receiving an error when I try to train a new Lora. The error message is: "An error occurred while downloading using `hf_transfer`. Consider disabling HF_HUB_ENABLE_HF_TRANSFER for better error handling." I've tried multiple ways to disable this including setting this variable to '0' in the terminal as well as with a python script and can't seem to get it to download the files without erroring out. I even nuked my pod and created a brand new one and am getting the same error. Any idea for how to disable this so I can continue my training?

Aitrepreneur

no but check that you are running the correct model, keep in mind that if you use other loras than lightx2v it might change the final result. You can also increase the weight of the lora as well

DFined

So, while the training seems to go great, the samples become good quite quickly, when I try to use the lora in the Wan t2i workflows, the result looks nothing like the samples - the effects are greatly diminished and often barely visible. For character loras it works fine, but for more complex concept loras - I keep running into this problem. Like a tried a battle-scarred-look lora and it worked great in the training samples, but in the workflows, the effects were much much weaker. I tried fiddling with samplers, cfg, etc, but still no luck. And the training itself seems to be working, as the samples do look good, but it just doesn't work in-workflow later. Do I need to train on a different model than just the base one or something?