open-thoughts

/

OpenThinker3-7B

like 116

Follow

Open Thoughts 598

Thank you!

It's fascinating to see how, while most are leaning heavily on RL-based fine-tuning, you've managed to achieve comparable results using just SFT. Impressive work!
Do you credit the dataset for this ?

Yep! The dataset is everything for us. We select a fairly standard training: SFT on Qwen2.5 :)