Which particular split on mesolitica/Malaysian-SFT dataset was this model trained on?
Hi! I'm trying to train Gemma-3-270m-it from Unsloth and I trained on the 'force_malay' split but the output is rubbish:
Here's the model I trained and published on HuggingFace if you're wondering. Appreciate the help!
the model is super small lora cant make it, you have to full parameter, 24GB VRAM should good enough.
the model is super small lora cant make it, you have to full parameter, 24GB VRAM should good enough.
So the 'force_malay' split should be ok is it? Just that I need to enable the full param fine-tuning? Or is there a better split to be used from the MalaysianSFT dataset
yep should be ok, you can add more force dataset to see how good the model scale along with the data.
yep should be ok, you can add more
forcedataset to see how good the model scale along with the data.
Awesome man, thanks for the guidance! I'll report back the findings for future reference 😁