arXiv:2606. 29869v1 Announce Type: cross Abstract: Knowledge distillation (KD) is a key technique for compressing Large Language Models (LLMs), yet methods relying on a single KL objective often fail to balance primary distribution fitting with long-tail probability modeling, limiting both generation quality and generalization.
Knowledge distillation (KD) is a key technique for compressing Large Language Models (LLMs), yet methods relying on a single KL objective often fail to balance primary distribution fitting with long-tail probability modeling, limiting both generation quality and generalization. To address this, we analyze the complementary roles of forward and reverse KL divergence (FKL/RKL) in distribution alignment from theoretical and empirical perspectives.
The paper introduces LLMR, a new knowledge distillation technique that uses a reward function derived from large language models to guide the distillation process. Experiments on dialogue generation and summarization datasets show that LLMR consistently outperforms traditional knowledge distillation methods across multiple tasks and datasets.
arXiv:2604. 00626v4 Announce Type: replace Abstract: As Large Language Models continue to grow in both capability and cost, transferring frontier capabilities into smaller, deployable students has become an important engineering problem, and knowledge distillation remains a common technique for this transfer.
arXiv:2605. 03677v2 Announce Type: replace Abstract: On-policy distillation (OPD) has recently emerged as an effective post-training paradigm for consolidating the capabilities of specialized expert models into a single student model.
arXiv:2604. 20244v2 Announce Type: replace-cross Abstract: Knowledge distillation (KD) is a powerful paradigm for compressing large language models (LLMs), whose effectiveness depends on intertwined choices of divergence direction, optimization strategy, and data regime.
arXiv:2607. 08776v1 Announce Type: cross Abstract: Despite the success of knowledge distillation (KD) in Large Language Models (LLMs), the underlying mechanism behind its efficacy remains unclear.
arXiv:2602. 22495v3 Announce Type: replace-cross Abstract: Reinforcement learning (RL) post-training has recently driven major gains in long chain-of-thought reasoning large language models (LLMs), but the high inference cost of such models motivates distillation into smaller students.
The paper introduces Round-Trip Reinforcement Learning (RTRL), a framework that trains chemical language models to improve round‑trip consistency by rewarding successful forward and reverse transformations. By iteratively training forward and reverse mappings, RTRL leverages abundant unlabeled chemical data to enhance both consistency and overall performance across supervised, self‑supervised, and synthetic data regimes. Experiments show that RTRL outperforms strong baselines, demonstrating that round‑trip consistency can be treated as a trainable objective for more robust foundation models.
PIVOT introduces a dynamic transition framework for vertical-domain few-shot distillation, routing samples between On-Policy Distillation (OPD) and GRPO-based reinforcement learning based on teacher-evaluated perplexity. Low-perplexity samples are moved to GRPO for reward-driven refinement, while high-perplexity samples remain under OPD to continue acquiring domain knowledge. Experiments on Banking77 and HWU64 demonstrate that PIVOT outperforms both continued OPD and globally synchronized OPD→GRPO baselines, achieving stronger downstream performance and more stable training dynamics.
The paper introduces Tail‑Corrected Top‑k On‑Policy Distillation (TT‑OPD), a method that improves on existing Top‑k OPD by combining the selected top‑k tokens with a sampled token from the student’s rollout. This hybrid approach recovers the probability mass discarded by limiting to top‑k, yielding an unbiased estimator of the reverse KL divergence gradient while maintaining low computational cost. Experiments show TT‑OPD outperforms other OPD variants in accuracy.
arXiv:2601. 07155v3 Announce Type: replace-cross Abstract: Knowledge distillation (KD) is a widely adopted technique for transferring knowledge from large language models to smaller student models; however, conventional supervised KD often suffers from a distribution mismatch between training and inference.