# Trying to train Loras for Illustrious locally — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/7f5123c6e70aeea6

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-10-29 07:09 ingested 2026-10-30 02:38

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1xid2b9/trying_to_train_loras_for_illustrious_locally/)

## retained text



```
First of all, I have a 1050Ti card and windows 10. I expect the training will be slow as it's not a particularly strong card but I don't mind waiting half a day or more if necessary.

Anyway, where do I even start?

I've tried installing two different training software. Neither worked. If you have the same card, what software worked for you?

I've tried these :

**Kohya SS.** This one gives me 

Error named symbol not found at line 177 in file D:\\a\\bitsandbytes\\bitsandbytes\\csrc\\ops.cu

Didn't find any useful information on how to fix that.

**Ostris Ai Toolkit**. This one gives me two errors : 

    1:
    r:\ai-toolkit\.venv\Lib\site-packages\torch\cuda\__init__.py:540: UserWarning: 

    NVIDIA GeForce GTX 1050 Ti with CUDA capability sm_61 is not compatible with the current PyTorch installation.r:\ai-toolkit\.venv\Lib\site-packages\torch\cuda\__init__.py:540: UserWarning: NVIDIA GeForce GTX 1050 Ti with CUDA capability sm_61 is not compatible with the current PyTorch installation.

    2 :
    UnicodeDecodeError: 'cp932' codec can't decode byte 0x91 in position 1: illegal multibyte sequence

    ... lots of text ... File "r:\ai-toolkit\.venv\Lib\site-packages\diffusers\models\model_loading_utils.py", line 208, in load_state_dict    raise OSError(OSError: Unable to load weights from checkpoint file for 'e:\AI\ED3\models\stable-diffusion\illustriousXL_v01.safetensors 

For the first one, I did uninstall the torch thing and reinstalled a different version that was recommended for this error message but it didn't change anything (maybe ai toolkit uses its own torch thingy instead of the global one on the system? It has a billion subfolders for all kinds of modules and plugins I think) and for the second error I had no useful search results. Not sure if it even needs fixing as it progresses beyond this point anyway, it's just a warning, not an error.

So it there some mythical third training software that is compatible with my GPU or something I can do to make one of these two work? As a temporal solution I've used civitai for a week but it's way too grindy, training costs several days worth of lighting bolt currency, and I didn't quit playing gacha games to start doing the same daily grind stuff elsewhere.

Can I even expect a simple character lora training of 20-50 images to finish in less than, idk, 16 hours on this GPU?
```



document fetched 2026-10-30 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1xid2b9/trying_to_train_loras_for_illustrious_locally
