Self-Policy Distillation extracts a capability subspace from model gradients on correctness tokens, projects KV activations into it for self-generation, and fine-tunes LLMs to achieve up to 13-16% gains over baselines without external signals.
hub
Embarrassingly simple self-distillation improves code generation
15 Pith papers cite this work. Polarity classification is still indexing.
15
Pith papers citing it
abstract
Can a large language model (LLM) improve at code generation using only its own raw outputs, without a verifier, a teacher model, or reinforcement learning? We answer in the affirmative with simple self-distillation (SSD): sample solutions from the model with certain temperature and truncation configurations, then fine-tune on those samples with standard supervised fine-tuning. SSD improves Qwen3-30B-Instruct from 42.4% to 55.3% pass@1 on LiveCodeBench v6, with gains concentrating on harder problems, and it generalizes across Qwen and Llama models at 4B, 8B, and 30B scale, including both instruct and thinking variants. To understand why such a simple method can work, we trace these gains to a precision-exploration conflict in LLM decoding and show that SSD reshapes token distributions in a context-dependent way, suppressing distractor tails where precision matters while preserving useful diversity where exploration matters. Taken together, SSD offers a complementary post-training direction for improving LLM code generation. Our code is available at https://github.com/apple/ml-ssd
hub tools
citation-role summary
background 1
citation-polarity summary
years
2026 15verdicts
UNVERDICTED 15roles
background 1polarities
background 1representative citing papers
EGRSD and CL-EGRSD advance the accuracy-length frontier in LLM reasoning by entropy-guided weighting of token-level distillation signals from the teacher.
PBSD derives a reward-reweighted teacher distribution as the analytic optimum of a reward-regularized objective, yielding better stability and performance than KL-based self-distillation on math reasoning and tool-use tasks.
SFR applies conditional flow matching on future sentence embeddings as a training regularizer to increase output diversity in style-conditioned LLMs without deployment overhead.
DASD improves math reasoning in LLMs by adaptively directing self-distillation based on per-token entropy to balance exploration and step accuracy, outperforming prior self-distillation and RLVR baselines on six benchmarks.
UniSD unifies self-distillation components for autoregressive LLMs and its full integrated version improves base models by 5.4 points and baselines by 2.8 points across six benchmarks.
LLMs exhibit Bayesian-like hypothesis updating with strong-sampling bias and an evaluation-generation gap but generalize poorly outside observed data.
The power distribution is the target of power sampling, the closed-form solution to self-reward KL-regularized RL, and the basis for power self-distillation that matches sampling performance at lower cost.
Iterative self-finetuning of LLMs mostly fails to amplify seeded behavioral traits, with amplification limited to specific DPO setups and often harming coherence.
SelfEvo enables pretrained 4D perception models to self-improve on unlabeled videos via self-distillation, delivering up to 36.5% relative gains in video depth estimation and 20.1% in camera estimation across eight benchmarks.
Controlled empirical study shows correcting Wikipedia data coverage yields larger gains than algorithm differences in LLM search agent training, with outcome-based rewards competitive.
AMR-SD adds a reflection bottleneck to compress diagnostic signals into self-generated hints and uses asymmetric Causal Information Gain to create sparse token-level advantage signals, outperforming baselines and preventing late-stage collapse in RLVR.
Nautile-370M is a hybrid small language model using SeqCond Attention layers alternating with transformers, with a claimed proof that the spectral operator matches full self-attention expressiveness in the continuous limit.
DASD dynamically selects tokens in self-distillation to keep logical corrections while suppressing stylistic noise, improving robustness on math, code, and commonsense benchmarks.
This overview paper explains the conceptual foundations and design principles of On-Policy Self-Distillation for large language models from a beginner's perspective.
citing papers explorer
-
Preference-Based Self-Distillation: Beyond KL Matching via Reward Regularization
PBSD derives a reward-reweighted teacher distribution as the analytic optimum of a reward-regularized objective, yielding better stability and performance than KL-based self-distillation on math reasoning and tool-use tasks.