C-DIC: Context-Driven Incremental Compression for efficient long-horizon multi-turn dialogue
A new arXiv preprint introduces Context-Driven Incremental Compression (C-DIC), a method for managing growing dialogue history in conversational agents by treating conversations as interleaved contextual threads with revisable per-thread compression states stored in a compact dialogue memory. A retrieve-revise-write-back loop shares information across turns and updates stale memories, while truncated backpropagation-through-time (TBPTT) is adapted to learn cross-turn dependencies. Experiments on long-form dialogue benchmarks show stable inference latency and perplexity over hundreds of turns, addressing compounding errors seen in existing context compressors.
Related guides (2)
Related events (8)
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.
COMPACT-VA: Planning-aligned token compression for long-context autonomous driving
Researchers introduce COMPACT-VA, a working memory framework using conditional VQ-VAE to compress extended temporal context in vision-action autonomous driving models. Compression is conditioned on historical trajectory and a learned planning intent derived from future trajectories during training, enabling end-to-end optimization without backbone modifications. On high-signal dynamic scenarios, the method achieves 68.3% success rate (>6% improvement) with 3.3x speedup and 2.7x memory reduction over uncompressed processing.
Canonical-Context On-Policy Distillation (CCOPD) for Multi-Turn LLM Consistency
This paper identifies 'self-anchored drift' as a key failure mode in multi-turn LLMs: when information is revealed incrementally across turns, models produce unsupported assumptions that distort final answers, even when the total evidence is identical to a single-prompt setting. The authors propose Canonical-Context On-Policy Distillation (CCOPD), which trains a student model on incremental multi-turn conversations to match the output distribution of a frozen teacher conditioned on the full clean prompt. Trained only on math conversations, CCOPD achieves a 32% average relative improvement on multi-turn (RAW-SHARDED) tasks and generalizes zero-shot to five out-of-domain task families while preserving single-prompt performance.
Information-theoretic metric for measuring semantic progress in multi-turn dialogue
A new arXiv preprint formalizes 'semantic progress' in multi-turn dialogue as question-conditioned uncertainty reduction and introduces an information-theoretic metric approximated in embedding space using a Gaussian formulation with closed-form updates. The metric has desirable theoretical properties (monotonicity, additive decomposition, diminishing returns) and requires no autoregressive inference at evaluation time, making it reproducible and lightweight. Experiments on MT-Bench, Chatbot Arena, and UltraFeedback show competitive or improved agreement with human judgments compared to several LLM-as-a-judge baselines. The approach works with lightweight embedding models under CPU-only execution.
Pair-In, Pair-Out (PIPO): Unified Latent Compression and Multi-Token Prediction for Efficient LLM Inference
PIPO is a new inference efficiency framework that unifies input-side latent compression with output-side multi-token prediction (MTP) by treating them as mirror operations: a compressor folds two input tokens into one latent, while an MTP head unfolds one hidden state into an additional output token. To avoid the expensive verifier pass typically required by speculative decoding, PIPO trains a lightweight confidence head using On-Policy Distillation (OPD), which naturally aligns with rejection-sampling criteria. Experiments on Qwen3.5-4B and 9B backbones across AIME 2025, GPQA-Diamond, LiveCodeBench v6, and LongBench v2 show up to 2.64× first-token-latency speedup and +7.15 pass@4 improvement over regular decoding.
TokenPilot: Dual-granularity context management cuts LLM agent inference costs by up to 87%
TokenPilot is a cache-efficient context management framework for LLM agents that addresses the trade-off between token sparsity and prompt cache continuity. It combines Ingestion-Aware Compaction (global prefix stabilization) with Lifecycle-Aware Eviction (local segment offloading) to reduce inference costs by 56–87% across benchmarks while maintaining competitive task performance. The system is evaluated on PinchBench and Claw-Eval and has been integrated into the open-source LightMem2 library.
Mastering Long Contexts in LLMs with KVPress
NVIDIA and Hugging Face present KVPress, a library for compressing the KV cache in large language models to enable more efficient long-context inference. The tool implements multiple KV cache compression ("pressing") algorithms that reduce memory footprint and latency without retraining models. KVPress is positioned as a practical toolkit for deploying LLMs in long-context scenarios where KV cache size becomes a bottleneck.
RL-based alignment improves interactivity in full-duplex spoken dialogue models
Researchers propose a post-training alignment method using reinforcement learning to improve interactivity in full-duplex spoken dialogue models, which can listen and speak simultaneously. The method addresses four canonical axes of interactivity—pause handling, turn-taking, backchanneling, and user interruption—each with axis-specific reward functions, plus an LLM-based reward to prevent semantic degradation. The approach is applied to two open-source models, Moshi and PersonaPlex, showing consistent improvements in both offline and real-time multi-turn evaluation.