Lightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation
cs.LG, cs.AI
Submitted: 2026-04-14
Updated: 2026-09-26
Code: https://github.com/jet-ai-projects/Lightning-OPD
License: http://creativecommons.org/licenses/by/4.0/
The gist: On-policy distillation (OPD) is an effective post-training paradigm for large language models but requires a live teacher server throughout training, resulting in substantial infrastructure overhead.
Terminology
Abstract
On-policy distillation (OPD) is an effective post-training paradigm for large language models but requires a live teacher server throughout training, resulting in substantial infrastructure overhead. We investigate whether OPD can be performed offline by precomputing teacher log-probabilities once over SFT rollouts and reusing them during training. We find that naively doing so fails to reliably match standard OPD, and trace the root cause to a previously overlooked condition we term teacher consistency, requiring that the same teacher be used for both supervised fine-tuning and OPD. Violating this condition introduces a gradient bias that degrades performance for both offline and online OPD. Building on this insight, we propose Lightning OPD, an offline on-policy distillation framework that enforces teacher consistency and eliminates the need for a live teacher server entirely. We prove that, under teacher consistency, Lightning OPD shares the same optimum as standard OPD, with bounded gradient discrepancy and an implicit regularization effect that helps prevent policy drift. Experiments on math reasoning and code generation show that Lightning OPD achieves comparable performance to standard OPD while delivering 4.0x higher training efficiency. Starting from an SFT-initialized Qwen3-8B-Base model, Lightning OPD reaches 69.9% on AIME 2024 in just 30 GPU hours. Lightning OPD further scales to MoE architectures, training Qwen3-30B-A3B to 71.0% on AIME 2024 on a single 8xH100 node, substantially lowering the barrier for academic research on LLM post-training. Our code is released at https://github.com/jet-ai-projects/Lightning-OPD.
Sources
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- OpenAI GPT-5 System Card
- Kimi K2.5: Visual Agentic Intelligence
- NVIDIA Nemotron 3: Efficient and Open Intelligence
- MiMo-V2-Flash Technical Report
- OpenThoughts: Data Recipes for Reasoning Models
- Learning beyond Teacher: Generalized On-Policy Distillation with Reward Extrapolation
- Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models
- Self-Distillation Enables Continual Learning
- Reinforcement Learning via Self-Distillation
- A Survey of On-Policy Distillation for Large Language Models
- Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe
- Revisiting On-Policy Distillation: Empirical Failure Modes and Simple Fixes
- Entropy-Aware On-Policy Distillation of Language Models
- Proximal Policy Optimization Algorithms
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- DAPO: An Open-Source LLM Reinforcement Learning System at Scale
- Qwen3 Technical Report
- Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?
- RL's Razor: Why Online Reinforcement Learning Forgets Less
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks