ADEPT: Accelerating Dexterity via Pre-Training and Post-Training using Reinforcement Learning — ThinkLLM