Interested to know more
Hi man, nice work 🔥
I am actually interested to know more about the process you followed to make this model ! I see you used axolotl for fine-tuning but what about vocab extension? Any particular layer were targeted? Continual pertaining? What about evaluating the model against the original Qwen1.5 ?
Again great job 👏🏻🤗
Hi Ali,
Thank you so much for finding my model interesting; it means a lot to me since I've only recently started getting into LLMs (you could definitely say I'm a noob 😁) , and this is my first fine-tuning attempt. Yes, I used Axolotl for training the model, and you can find the details of the fine-tuning process in the YAML file. This model, along with the new one I uploaded today, is more about learning the finetuning process for me. I hope to improve myself as well. I've just learned what continual pre-training is. Also, I'm not sure how to compare the model's success on Turkish tasks with the original model, but this fine-tune seems to perform much better compared to other Turkish fine-tunes on Hugging Face - though this is likely more due to Qwen's models being the best open-source model with Turkish capabilities rather than my fine-tuning effort.
By the way, thanks for the encouragement, my man ❤️🫡