We released the first successful replication of DeepSeek-R1's 'aha moment' in a multimodal task using only a 2B non-SFT model!

We released the first successful replication of DeepSeek-R1’s ‘aha moment’ in a multimodal task using only a 2B non-SFT model!