Yihua Zhang
EN 中

Articles

All13 Post-training6 Systems5 Music2 Archive20

Series

3

Anatomy of DeepSeek

 3

The GRPO lineage

13 articles

Vogel im Käfig: How Sawano Builds Hope and Despair from the Smallest Interval
Music · zh Sawano
A Role Shift for AI Infra: From Foundational Support to a Core Engine of Innovation

AI Market Insights

Systems · en
From GRPO to DAPO and GSPO: What, Why and How
Post-training · en / zh ·18 equations The GRPO lineage 2 GRPO
Re-understanding KL Approximation from an RL-for-LLM Lens: Notes on “Approximating KL Divergence
Post-training · en / zh ·18 equations The GRPO lineage 3 GRPO
Decorators in Machine Learning Projects
Systems · en / zh Python
Bauklötze: A Musical Dissection

积木崩塌时的命运回响，泽野弘之用音符砌筑的巨人悲歌

Music · zh Sawano
DualPipe Explained: A Comprehensive Guide to DualPipe That Anyone Can Understand—Even Without a Distributed Background
Systems · en / zh Anatomy of DeepSeek 3 Distributed
Navigating the RLHF Landscape: From Policy Gradients to PPO, GAE, and DPO for LLM Alignment
Post-training · en / zh ·88 equations RLHF
DeepSeek-R1 Dissection: Understanding PPO & GRPO Without Any Prior Reinforcement Learning Knowledge
Post-training · en / zh ·16 equations The GRPO lineage 1 GRPORLHF
Why Cache 32 Heads When One Latent Variable Suffices? A Theory-to-Code Guide to DeepSeek’s MLA for KV-Cache
Systems · en / zh ·24 equations Anatomy of DeepSeek 2 KV-Cache
From Zero to Reasoning Hero: How DeepSeek-R1 Leverages Reinforcement Learning to Master Complex Reasoning
Post-training · en / zh ·2 equations Anatomy of DeepSeek 1 GRPO
A Review on the Evolvement of Load Balancing Strategy in MoE LLMs: Pitfalls and Lessons
Systems · en / zh ·31 equations MoE
Patching the Foundation Models: Pitfalls and Pains in Machine Unlearning
Post-training · en / zh ·14 equations Unlearning