FireGate: Bounding Behaviour Change Before Training in Continual RL Fine-Tuning
cs.LG, cs.AI
Submitted: 2026-11-09
Updated: 2026-11-09
Code: https://github.com/mahemon/firegate
Terminology
Sources
- Unforgotten Safety: Preserving Safety Alignment of Large Language Models with Continual Learning
- Program Synthesis with Large Language Models
- Pay Only for Disagreement: Certified No-Regression Verdicts for Model Updates with Matching Label-Complexity Bounds
- DeepSeek LLM: Scaling Open-Source Language Models with Longtermism
- Fight Fire with Fire: Defending Against Malicious RL Fine-Tuning via Reward Neutralization
- Fundamental Safety-Capability Trade-offs in Fine-tuning Large Language Models
- Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
- Training Verifiers to Solve Math Word Problems
- Token Buncher: Shielding LLMs from Harmful Reinforcement Learning Fine-Tuning
- The Llama 3 Herd of Models
- STABLE: Gated Continual Learning for Large Language Models
- Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey
- Mistral 7B
- Defending MoE LLMs against Harmful Fine-Tuning via Safety Routing Alignment
- Reinforcement Fine-Tuning Naturally Mitigates Forgetting in Continual Post-Training
- Examining Forgetting in Continual Pre-training of Aligned Large Language Models
- Beyond Forgetting: Diagnosing and Harnessing Shared Reasoning in Continual RLVR
- RL Forgets! Towards Continual Policy Optimization
- When Can Old Evaluations Certify a New Model? Label-Efficient Release Decisions under Evaluator Drift
- Preventing Catastrophic Forgetting: Behavior-Aware Sampling for Safer Language Model Fine-Tuning
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis