Reward Function As A Service - A (Relatively) Easy Recipe for Training Your Own Reasoning Model

Learn to post-train large language models using custom reward functions in this 35-minute conference talk from MLOps World. Discover a practical template for customizing high-performing LLMs like Deepseek for domains that support programmatic evaluation through Python functions. Explore how this approach differs from traditional supervised fine-tuning and extends beyond typical applications in math and coding to various other domains. Follow along as the speaker demonstrates the implementation process and provides creative examples of applying reward function-based training to customize reasoning models for specific use cases. Gain insights into making advanced LLM customization more accessible through developer-friendly approaches and practical templates that can be adapted for real-world ML/AI systems.