The entire training process of DeepSeek R1 is nothing but using different way of reinforcement learning on top of their base model (i.e. deepseek V3)
Starting with a tiny base model that runs locally, we’ll build everything from scratch using DeepSeek R1 tech report while covering theory alongside each step.
Whether you are new to DeepSeek R1 or want to train your own model, this blog has you covered! 🚀
This article visually explains how DeepSeek R1 works, from start to finish. Read it here:
After SFT, the model can reason better, but we want to really focus on reasoning quality and fix language mixing. This stage uses RL again, but with a smarter reward system.
That makes sense! After Supervised Fine-Tuning (SFT), reinforcement learning (RL) can refine reasoning quality by optimizing for more precise and coherent outputs. A smarter reward system ensures the model improves in the right direction—whether…
Which platform do you use Deep Seek on? I enjoyed your article. So useful. I used it on Galaxy platform and am now keen to use it in Linux. I’m looking to get more functionality from it.
