Qwen/Qwen2.5-0.5B, but with the vocab of deepseek-ai/DeepSeek-R1-Distill-Qwen-32B transplanted using transplant-vocab.
Base model