Hey!
So I get a ton of errors when I try to train CPU only on ez trainer, even though I have a correctly formatted json file. Any tips for how one might get into training on the shallow end?
Trynna get a dataset I've made into trained into Synthia7b 1.3 (the mistral version of Synthia). Fully merged, no lora adapter. It's long format stories broken into 140 chunks. Tough to work out if you aren't a python fellow!
All this stuff reminds me of writing batch files. Wish ML was easier lol! Thanks for your time.
I'm unfamiliar with how ez trainer is used, so unfortunately, I cannot help with this as I use Axolotl. Axolotl has more settings regarding what can be done with it but it's also more difficult to set up (additionally, you'll have to use WSL if you're attempting it on Windows).
I would recommend asking for help in TheBloke's Discord server regarding how to use ez trainer as it seems there are people who have used it but I can provide some help with my basic Python knowledge.
Thanks I'll give asking in there a go :)