The Gradient That Top-k Distillation Never Computes

arXiv:2610.15589 · cs.LG, cs.CL · Submitted 2026-10-11 · Read on arXiv

cs.LG, cs.CL

Submitted: 2026-10-11

Updated: 2026-10-11

Code: https://github.com/KurbanIntelligenceLab/top-k-distillation

Terminology

• Distillation — 1× in this paper · explained in 2 episodes

Sources

• Distillation Scaling Laws
• Stealing Part of a Production Language Model
• DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence
• Logits of API-Protected LLMs Leak Proprietary Information
• Representation Degeneration Problem in Training Natural Language Generation Models
• Why do small language models underperform? Studying Language Model Saturation via the Softmax Bottleneck
• Distilling the Knowledge in a Neural Network
• Liger Kernel: Efficient Triton Kernels for LLM Training
• The Geometry of Last-Layer Model Stealing
• DistiLLM: Towards Streamlined Distillation for Large Language Models
• DistiLLM-2: A Contrastive Approach Boosts the Distillation of LLMs
• On the Generalization of Knowledge Distillation: An Information-Theoretic View
• Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe
• Your Teacher Can't Help You Here: Combating Supervision Fidelity Decay in On-Policy Distillation
• FitNets: Hints for Thin Deep Nets
• Efficient Knowledge Distillation for LLMs: Offline Top-K Logits and a Fused Chunked KL Loss
• Geometric Limits of Knowledge Distillation: A Minimum-Width Theorem via Superposition Theory
• A Survey of On-Policy Distillation for Large Language Models
• The Illusion of Certainty: Decoupling Capability and Calibration in On-Policy Distillation
• KDFlow: A User-Friendly and Efficient Knowledge Distillation Framework for Large Language Models

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