Tony Salomone
Tony Salomone
Possibly need to file bug with MLX.
I suspect this is a timeout and we just aren't handling well. The mixtral model, even at 4bit, is about 26GB vs. Phi 3 4-bit at about 8 GB. On...
Still waiting for this fix in electron. Added a note to the docs about --no-sandbox for now.
The MLX LoRA trainer plugin takes a very simple approach and essentially doesn't handle any formatting for you. i.e. It does not convert your JSON into a chat template that...
Hmmm that is peculiar. is the pad string on this model so it shouldn't display that. But what's weirder is that you didn't get a response before that. It's hard...
No! I am not sure when they added that, but I will look into this and fix! Thanks for finding this!
Hmmm...looks like we need something somewhere to catch this and return as an error so the app can display this to the user. Alternatively...the app should probably know when it's...
This fix was included in the last release (v0.18.0 released on Monday). Closing but please reopen if you find any issues!
Sorry my video didn't post but the dataset we were trying to download was `openai/gsm8K`