Long time ago (actually only a year ago), DeepSeek released a few open source model, such as deepseek-r1-distill-qwen (https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B).
I am wondering if anyone in the community is brave enough to make a DeepSeek-v4-distall-Qwen3.6-27b.
It would be really interesting to know if the distillation of DeepSeek can improve qwen3.6-27b further. The open-source deepseek-v4 can give us the internal data for distillation, unlike closed-source models.
[link] [comments]