I don't know where you're getting your information from. Maybe you're confusing DeepSeek v3/r1 and the distilled r1 models. DeepSeek V3/R1 architecture isn't a

I don't know where you're getting your information from. Maybe you're confusing DeepSeek v3/r1 and the distilled r1 models. DeepSeek V3/R1 architecture isn't anything like Llama 3. Llama 3 isn't even a mixture of experts, not to mention the various other differences like attention compression etc