Deepseek-R1-Distilled-Qwen-1.5B

UC伯克利团队通过RL微调改进Deepseek-R1-Distilled-Qwen-1.5B，使其在AIME基准上Pass@1准确率高达43.1%，参数量仅为1.5B且超越OpenAI o1-preview。