Heads up on the Deepseek model fine tuning for Sophia
After generating tenths of test datasets and trying to make the training work using free tier on hugging face, I am close to get the DeepSeek-R1-Distill-Qwen-1.5B model to be fine tuned on Sophia, but now the “no hardware free tier constraint" does not allow me to progress anymore, so I am gonna get some GPU power to continue this endeavour.
Once I get the proper config and dataset format for the training to make it work, I will start sharing datasets here for public scrutiny and feedback / contribution… a model is just as good as its dataset.
If there is any hugging face worm that feels like helping out here ping me please !