DEX-Comp combines distillation on teacher-correct examples with reinforcement learning on teacher-failed examples to produce soft context compression that surpasses uncompressed RAG.
Title resolution pending
6 Pith papers cite this work, alongside 6 external citations. Polarity classification is still indexing.
6
Pith papers citing it
6
external citations · Crossref
years
2026 6representative citing papers
Presents the first Õ(log^{1.5} n)-approximation algorithm for the graph label selection problem under a standard budget constraint.
An evolution-searched set of deterministic linguistic rules compresses LLM prompts without any language-model scoring at deployment and stays competitive on longer-document QA tasks.
CRAFT is a Pareto-front prompt optimizer that allocates scarce LLM validation calls to candidates near the current front using accuracy- and cost-oriented generators plus NSGA-II retention.
AttnCompress achieves 53.17% pass rate on SWE-Bench-Verified, outperforming prior compression methods while reducing tokens by 21.6% and total costs by 33.6%.
citing papers explorer
-
Compression Beyond the Uncompressed: A Two-Stage Training Recipe for Soft Context Compression in RAG
DEX-Comp combines distillation on teacher-correct examples with reinforcement learning on teacher-failed examples to produce soft context compression that surpasses uncompressed RAG.
-
An Approximation Algorithm for Graph Label Selection
Presents the first Õ(log^{1.5} n)-approximation algorithm for the graph label selection problem under a standard budget constraint.
-
Every Time I Hire a Linguist, Inference Costs Go Down: On Linguistic Rules as Effective Prompt Compressors
An evolution-searched set of deterministic linguistic rules compresses LLM prompts without any language-model scoring at deployment and stays competitive on longer-document QA tasks.
-
CRAFT: Cost-aware Refinement And Front-aware Tuning of Prompts
CRAFT is a Pareto-front prompt optimizer that allocates scarce LLM validation calls to candidates near the current front using accuracy- and cost-oriented generators plus NSGA-II retention.
-
AttnCompress: Dynamic Attention-Guided Trajectory Compression for Software Engineering Agents
AttnCompress achieves 53.17% pass rate on SWE-Bench-Verified, outperforming prior compression methods while reducing tokens by 21.6% and total costs by 33.6%.
- SALT: Salience-Aware Lexical Trie for Long-Context Compression