DeepSeek-R1

[Full Workshop] Reinforcement Learning, Kernels, Reasoning, Quantization & Agents — Daniel Han

Jul 19, 2025 · 2:42:28

Daniel Han of Unsloth presents a technical workshop covering reinforcement learning (RL), kernels, reasoning, quantization, and agents, arguing that RL with verifiable rewards (RLVR) is the key to unlocking LLM capabilities beyond supervised fine-tuning. He explains why open-source models plateaued after September 2024 until DeepSeek-R1 showed that RL can elicit reasoning, and breaks down PPO, GRPO, and the REINFORCE algorithm, emphasizing that GRPO removes the value model for efficiency. Han details how reward functions—not algorithms—are the hardest part, with examples like distance-based scoring for math. He demonstrates a free Colab notebook training a base model to reason, and shows that dynamic quantization can shrink models like DeepSeek-R1 from 730 GB to 140 GB with only ~1% accuracy loss, arguing that GPUs may stop getting faster after FP4 precision.

AI Agents, Meet Test Driven Development

Feb 22, 2025 · 29:10

Anita from Vellum argues that teams using test-driven development build more reliable AI systems, and she breaks down how to apply TDD to agentic workflows across four stages: experiment, evaluate, deploy, monitor. She outlines five levels of agentic behavior (L0 to L4), noting that most production systems today are at L1 (tool use), while L2 systems that plan and reason with models like O1 and DeepSeek-R1 will see most innovation this year. She introduces Vellum's new open-source Workflows SDK, which keeps code and UI in sync, and demonstrates her own SEO agent that automates keyword research, content analysis, and writing via a writer-editor evaluation loop. The agent, built on Vellum Workflows, takes a keyword like "chain-of-thought prompting" and produces a first draft with a latency of around 118 seconds, incorporating competitive analysis and iterative feedback. She emphasizes that success now depends on orchestration techniques (prompt chaining, RAG, memory) rather than just model performance.