Rethinking Training-Inference Mismatch in LLM Reinforcement Learning: Where It Arises and How to Correct It
Paper • 2609.32444 • Published • 24
🔄 In a Training Loop
Joel Wang
joelhenwang
AI & ML interests
None yet
Recent Activity
View all activity
Organizations
upvoted 20
papers 1 day ago
How Reproducible Are Evaluation Conclusions? A Self-Audit of LLM-Inferred Prompt Structure
Paper • 2609.30074 • Published • 2
Adaptive Consistency Graph for Long-Horizon Agents
Paper • 2609.32754 • Published • 3
Self-Play Search Distillation for Large Language Model Reasoning
Paper • 2609.30936 • Published • 4
WhiteMatter: All-to-All Cross-Layer Connections via KV Source Mixing
Paper • 2608.18486 • Published • 4
Allspark: Weak to Strong Transfer via Alternating Chain of Thought
Paper • 2609.32913 • Published • 4
Not All Objectives Are Born Equal: Priority-Constrained Descent for Hierarchical Multi-Objective Optimization
Paper • 2606.29521 • Published • 5
G^2PTQ: Improving LLM Post-Training Quantization with Generalized Gradient Compensation
Paper • 2609.31009 • Published • 4
Reinforcing Agentic Creativity in Scientific Ideation with Night Science
Paper • 2609.35706 • Published • 4
Who Gets a Token, and What Does It Carry? Unequal Name Support and Concept Access in Large Language Models
Paper • 2609.34065 • Published • 3
Approximating Softmax in Pretrained LLMs: Model Sensitivity and Kernel Acceleration
Paper • 2609.33586 • Published • 4
KernelZero: Co-Evolving Proposer and Coder for Continuously Improved GPU Kernel Generation
Paper • 2609.33074 • Published • 5
SMAT: Simple and Efficient Merge-Aware Training
Paper • 2609.33437 • Published • 5
FactorEngram: Factorized N-gram Memory with Basis-Level Gating for Language Models
Paper • 2609.35578 • Published • 4
WaveFront Decoding: Parallelized Self-Speculative Decoding for Looped Language Models
Paper • 2609.23033 • Published • 7
Can We Trust the Teacher? Decoupled Credit Direction-Magnitude for Self-Distillation
Paper • 2609.34848 • Published • 6
Fewer Tokens, More Self-Teaching: On-Policy Self-Distillation for Extreme Visual Token Reduction
Paper • 2609.32353 • Published • 8
ExpVoyager: Direct Experience Navigation for Dynamic Agent Skill Synthesis
Paper • 2609.32630 • Published • 14
Nereus: Adaptive Parallelism for LLM Post-Training
Paper • 2609.34645 • Published • 18
Change the Product, Keep the Parameters: Associative Algebra Layers for Transformers
Paper • 2609.32814 • Published • 21