SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.03784 · 2026

Fix the Structural Bottleneck: Context Compression via Explicit Information Transmission

Heng Chang, Ye Mao, Hanqi Yan, Jiangnan Ye, Zhenyi Shen, Mitigates Dilution

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 5 functions out of this paper's own repositories and ran 0 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
Jiangnan0522/ComprExIT canonical 0 of 5
FunctionStatusWhere it lives
beacon_predict Not yet run Jiangnan0522/ComprExIT/src/model/inference.py
code served (permissive licence) · get_code("6cfd975f730cf408")
data_loading_factory Not yet run Jiangnan0522/ComprExIT/src/data_processing/data_loading.py
code served (permissive licence) · get_code("9f88a25274d07db2")
get_data_collator_factory Not yet run Jiangnan0522/ComprExIT/src/data_processing/ntp_preprocessing.py
code served (permissive licence) · get_code("1d56830e39983466")
get_device_stats Not yet run Jiangnan0522/ComprExIT/src/device_utils.py
code served (permissive licence) · get_code("392995da7f414f44")
supports_flash_attention_2 Not yet run Jiangnan0522/ComprExIT/src/device_utils.py
code served (permissive licence) · get_code("414973f82aff6b26")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Long-context LLM agents often struggle with growing token, memory, and latency costs, making efficient context compression essential for practical deployment. Existing LLM-as-a-compressor methods remain noticeably inferior to using the full context. We find that this gap partly stems from their inability to preserve contextual information effectively. In this work, we revisit context compression from a structural perspective and identify two key bottlenecks in standard LLM-based compressors: limited coordination among compression tokens during information aggregation, and layerwise dilution that weakens useful signals from intermediate hidden states. To address these limitations, we propose Com-prExIT (Context Compression via Explicit Information Transmission), a new context compression framework based on explicit information transmission. Com-prExIT adaptively aggregates features into anchors across frozen LLM layers, then allocates information from anchors to compression slots through a globally coordinated transport plan. Experiments on 12 datasets show that ComprExIT consistently outperforms strong soft-compression baselines, improving average F1 by up to 18.5%, while adding only ∼1% trainable parameters and achieving over 2× faster compression than the fastest baselines. Code is available at https://github.com/Jiangnan0522/ComprExIT.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2602.03784")
get_code_for_paper("2602.03784")
have("2602.03784")

Connect an agent — have() is free.