7arXiv cs.LG (Machine Learning)·
Requential coding: self-generated training data enables orders-of-magnitude model compression and tighter generalization bounds
A new arXiv preprint introduces requential coding, a compression scheme where a teacher model selects training samples from the student model's own distribution, so the code records only teacher-student disagreements rather than the full data sequence. The resulting code lengths are independent of both parameter count and data entropy, yielding codes orders of magnitude shorter than prequential coding baselines. Plugged into a PAC-Bayes framework, the method produces state-of-the-art generalization guarantees for billion-parameter LLMs that tighten with scale in the compute-optimal regime, and reveals that larger models and ensembles become increasingly compressible relative to dataset size.
Related guides (2)
Related events (8)
Prompt compression via learned activation aggregation reduces per-query compute with under 2% accuracy drop
A new arXiv preprint proposes compressing instruction prompts into a single activation vector by taking a learned weighted sum of intermediate-layer activations and re-injecting it at an early layer of the target LLM. The method achieves under 2% accuracy degradation relative to full prompt processing, eliminating the need to reprocess fixed instruction prompts on every query. The work also surfaces structural findings about LLM activation spaces: mid-layer representations transfer to early layers, and a single vector can encode recoverable semantic information.
ReuseRL: Skill Reuse as Compression in Agentic RL via MDL Principle
ReuseRL formalizes agentic reinforcement learning through the Minimum Description Length (MDL) principle, extracting a shared skill dictionary from successful trajectories and augmenting the RL objective with a segmentation cost that penalizes idiosyncratic, non-reusable behaviors. The authors prove a PAC-Bayes generalization bound for this compression penalty. Evaluated on ALFWorld, TextWorld-Cooking, and Countdown-Stepwise, ReuseRL outperforms vanilla GRPO and round-length baselines on both in-distribution and out-of-distribution tasks.
CHERRY: Compressed Hierarchical Experts with Recurrent Representational Yield — three compute-efficient LM training techniques
A preprint from arXiv introduces CHERRY, a suite of three complementary techniques for compute-efficient language model training: Selective Ground Truth Token Training (SGT) that concentrates supervision on ~15% of semantically loaded tokens while recovering ~67% of full-sequence loss reduction; depth compression that shrinks a 48-layer 1B-parameter model to 6 layers (227M) via layer averaging and recurrent unrolling, matching a 566M dense model's loss; and a Mixture of Efficient Experts (MoEE) assembly that outperforms individual compressed models at comparable active parameters. The techniques are validated on CHERRY-1.8B, a Korean-language foundation model trained entirely from scratch using these methods. Authors are transparent about scope limitations: one model family, Korean data, and loss-based metrics only.
HiReLC: Hierarchical Reinforcement Learning Framework for Joint Neural Network Pruning and Quantization
Researchers introduce HiReLC, a hierarchical ensemble-RL framework that automates joint quantization and structured pruning of deep neural networks. The system uses two-level agents — low-level agents selecting per-kernel compression configurations and high-level agents coordinating global budget allocation via Fisher Information-based sensitivity estimates. Experiments on Vision Transformers and CNNs achieve 5.99–6.72× parameter-storage compression with accuracy drops of 0.55–5.62% in most settings. The controller is architecture-agnostic, using a surrogate MLP and active learning loop to reduce policy evaluation cost.
Randomized YaRN improves LLM length generalization for long-context reasoning
Researchers propose Randomized YaRN, a training method that combines YaRN-based positional extrapolation with randomized positional encodings and a length curriculum to improve LLM generalization to long contexts. Models trained on sequences under 8K tokens show consistent reasoning improvements on context lengths from 16K to 128K on BABILong and MRCR benchmarks. The key insight is that exposing models to out-of-distribution positional representations during short-context training enables better generalization at far longer inference-time lengths.
ReContext: Training-free recursive evidence replay improves LLM long-context reasoning
Researchers introduce RECONTEXT, a training-free inference-time method for improving long-context reasoning in LLMs. The approach uses model-internal relevance signals to build a query-conditioned evidence pool that is replayed before final generation, without modifying the original context, external memory, or context pruning. Experiments across eight long-context datasets at 128K context length show consistent improvements on Qwen3-4B, Qwen3-8B, and Llama3-8B. The authors provide a theoretical grounding via associative memory theory, framing attention as cue-trace association and replay as trace reactivation.
Latent Context Language Models (LCLMs) achieve competitive encoder-decoder KV cache compression at scale
Researchers introduce Latent Context Language Models (LCLMs), a family of encoder-decoder compressors that map long token sequences to shorter latent embeddings consumed by a decoder, targeting the KV cache memory bottleneck in long-context inference. The authors conduct architecture search and continually pre-train 0.6B-encoder/4B-decoder models on over 350B tokens at compression ratios of 1:4, 1:8, and 1:16. LCLMs improve the Pareto frontier across general-task performance, compression speed, and peak memory, and are demonstrated as efficient backbones for long-horizon agents that can skim compressed context and expand relevant segments on demand. The work closes a previously noted gap between encoder-decoder approaches and KV cache compression methods on the accuracy-efficiency frontier.
CoRP: Gradient-Free Consolidation of Rewarded Perturbations for LLM Post-Training
CoRP (Consolidating Rewarded Perturbations) is a gradient-free post-training operator that folds an ensemble of reward-weighted weight-space perturbations into a single deployable model, eliminating the inference-time cost of ensemble methods like RandOpt. A split-half analysis across 25 model-task pairs reveals reproducible low-rank structure in the rewarded perturbation population, which CoRP exploits via reward-weighted aggregation, compatibility-aware reweighting, and a held-out validation gate. Evaluated on five models (0.5B–8B) across math, code, and creative writing, CoRP improves the base model by 8.1 points on average, exceeds single-inference RandOpt by 6.5 points using one-tenth the perturbation budget, and recovers more than half the gain of a 50-pass majority-vote ensemble at one forward pass per test example.