DistiLLM: Towards Streamlined Distillation for Large Language Models
read the original abstract
Knowledge distillation (KD) is widely used for compressing a teacher model to a smaller student model, reducing its inference cost and memory footprint while preserving model capabilities. However, current KD methods for auto-regressive sequence models (e.g., large language models) suffer from missing a standardized objective function. Moreover, the recent use of student-generated outputs to address training-inference mismatches has significantly escalated computational costs. To tackle these issues, we introduce DistiLLM, a more effective and efficient KD framework for auto-regressive language models. DistiLLM comprises two components: (1) a novel skew Kullback-Leibler divergence loss, where we unveil and leverage its theoretical properties, and (2) an adaptive off-policy approach designed to enhance the efficiency in utilizing student-generated outputs. Extensive experiments, including instruction-following tasks, demonstrate the effectiveness of DistiLLM in building high-performing student models while achieving up to 4.3$\times$ speedup compared to recent KD methods.
This paper has not been read by Pith yet.
Forward citations
Cited by 17 Pith papers
-
Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients
ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite w...
-
Bounded Behavioral Indistinguishability for Black-Box LLM Distillation
Introduces (ε,q,t,A)-behavioral indistinguishability and shows via Qwen/Llama experiments that LoRA distillation boosts semantic similarity but leaves detectable behavioral differences under adversarial evaluation.
-
Visual-Advantage On-Policy Distillation for Vision-Language Models
VA-OPD improves VLM performance over standard on-policy distillation by reweighting rollouts and separating KL terms according to token-level visual advantage on math and visual benchmarks.
-
The Extrapolation Cliff in On-Policy Distillation of Near-Deterministic Structured Outputs
On-policy distillation has an extrapolation cliff at closed-form lambda*(p,b,c) set by teacher modal probability, warm-start mass, and clip strength, past which training shifts from format-preserving to format-collapsing.
-
MTA: Multi-Granular Trajectory Alignment for Large Language Model Distillation
MTA improves LLM knowledge distillation by aligning representations along layer-wise trajectories with adaptive granularity from words to phrases using dynamic structural and hidden representation alignment losses.
-
PHF: Privileged Hidden Flow for On-Policy Self-Distillation
PHF distills token-to-token transition directions and trajectory geometry in hidden states during on-policy self-distillation, reporting 1.5-2.2 point gains on Average@12 for Qwen3-1.7B/4B/8B over reproduced OPSD base...
-
RLCSD: Reinforcement Learning with Contrastive On-Policy Self-Distillation
RLCSD contrasts teacher-student distributional gaps under correct versus wrong hints to suppress privilege-induced style drift and concentrate supervision on task tokens, outperforming GRPO and prior OPSD on Qwen3 and...
-
Contribution Weights: A Geometrical Analysis of Self-Attention Transformers
Contribution Weights combine attention, value magnitude, and directional alignment to measure token influence more faithfully than attention alone, and show attention sinks actively suppress information via a convex s...
-
When Are Teacher Tokens Reliable? Position-Weighted On-Policy Self-Distillation for Reasoning
Position-Weighted On-Policy Self-Distillation (PW-OPSD) weights later tokens more heavily after a diagnostic shows position predicts teacher reliability better than entropy, yielding +1.0 and +1.1 Avg@12 gains on AIME...
-
Learning with Rare Success but Rich Feedback via Reflection-Enhanced Self-Distillation
RESD turns failure trajectories into token-level supervision via retrospective reflections and a persistent global playbook, enabling faster improvement than standard self-distillation or GRPO with only one rollout pe...
-
Hybrid Policy Distillation for LLMs
Hybrid Policy Distillation unifies existing knowledge distillation methods for LLMs into a reweighted log-likelihood objective and introduces a hybrid forward-reverse KL approach with mixed data sampling to improve st...
-
Switch-KD: Visual-Switch Knowledge Distillation for Vision-Language Models
A 0.5B student VLM distills from a 3B teacher using visual-switch distillation and DBiLD loss to gain 3.6 points on average across 10 multimodal benchmarks without architecture changes.
-
PriFT: Prior-Support Guided Supervised Fine-Tuning
PriFT uses token reweighting signals from a frozen pretrained model to stabilize SFT and achieve better results than standard SFT baselines on reasoning tasks.
-
MoASE++: Mixture of Activation Sparsity Experts with Domain-Adaptive On-policy Distillation for Continual Test Time Adaptation
MoASE++ combines activation sparsity experts with domain-adaptive on-policy distillation to achieve state-of-the-art continual test-time adaptation on image classification and segmentation benchmarks.
-
Curriculum Learning-Guided Progressive Distillation in Large Language Models
CLPD improves LLM distillation for reasoning by combining explicit data curriculum with progressive teacher scheduling of increasing capacity.
-
Near-Policy: Accelerating On-Policy Distillation via Asynchronous Generation and Selective Packing
NPD accelerates on-policy distillation 8.1 times faster than baselines by using asynchronous SFT with Δ-IFD filtering, outperforming standard SFT and enabling a 1B model to achieve 68.73% SOTA score.
-
MTA: Multi-Granular Trajectory Alignment for Large Language Model Distillation
MTA is a distillation method that aligns teacher-student LLM representations along their transformation trajectories using layer-adaptive granularities and dynamic structural plus hidden representation alignment losses.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.