Enhancing Fitness Intelligence through Domain-Specific LLM Post-Training

FitOne optimizes fitness coaching via a three-stage post-training pipeline including continual pre-training, supervised fine-tuning, and reinforcement learning. The 8B and 32B parameter models, built on Qwen3, utilize high-quality knowledge engineering to address domain-specific gaps in general-purpose LLMs for complex scientific fitness coaching scenarios.