Pith. sign in

REVIEW 7 cited by

QwenLong-CPRS: Towards infty-LLMs with Dynamic Context Optimization

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2505.18092 v2 pith:45UQKP2N submitted 2025-05-23 cs.CL

QwenLong-CPRS: Towards infty-LLMs with Dynamic Context Optimization

classification cs.CL
keywords contextqwenlong-cprsllmsoptimizationperformancecompressiondynamiclanguage
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

This technical report presents QwenLong-CPRS, a context compression framework designed for explicit long-context optimization, addressing prohibitive computation overhead during the prefill stage and the "lost in the middle" performance degradation of large language models (LLMs) during long sequence processing. Implemented through a novel dynamic context optimization mechanism, QwenLong-CPRS enables multi-granularity context compression guided by natural language instructions, achieving both efficiency gains and improved performance. Evolved from the Qwen architecture series, QwenLong-CPRS introduces four key innovations: (1) Natural language-guided dynamic optimization, (2) Bidirectional reasoning layers for enhanced boundary awareness, (3) Token critic mechanisms with language modeling heads, and (4) Window-parallel inference. Comprehensive evaluations across five benchmarks (4K-2M word contexts) demonstrate QwenLong-CPRS's threefold effectiveness: (1) Consistent superiority over other context management methods like RAG and sparse attention in both accuracy and efficiency. (2) Architecture-agnostic integration with all flagship LLMs, including GPT-4o, Gemini2.0-pro, Claude3.7-sonnet, DeepSeek-v3, and Qwen2.5-max, achieves 21.59$\times$ context compression alongside 19.15-point average performance gains; (3) Deployed with Qwen2.5-32B-Instruct, QwenLong-CPRS surpasses leading proprietary LLMs by 4.85 and 10.88 points on Ruler-128K and InfiniteBench, establishing new SOTA performance.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Do Generative AI Assistants Respect robots.txt? Tracing Web Access Beyond Visible Answers

    cs.CY 2026-07 conditional novelty 6.0

    Under controlled tests, several major AI assistants (Gemini, Grok, DeepSeek, Qwen) accessed robots.txt-disallowed web pages without ever requesting the robots.txt file, while Claude and Mistral generally complied.

  2. ECHO: Prune To Act, Trace To Learn With Selective Turn Memory In Agentic RL

    cs.LG 2026-06 unverdicted novelty 6.0

    ECHO is a selective turn-memory framework for agentic RL that compresses turns into indexed records, selects them for bounded contexts, and uses source indices to assign outcome credit to supporting evidence, reaching...

  3. ECHO: Prune To Act, Trace To Learn With Selective Turn Memory In Agentic RL

    cs.LG 2026-06 conditional novelty 6.0

    Source-indexed selective turn memory plus provenance-guided credit assignment raises BrowseComp-Plus accuracy to 43.4% over GRPO and SUPO while cutting turn proliferation.

  4. ECHO: Prune To Act, Trace To Learn With Selective Turn Memory In Agentic RL

    cs.LG 2026-06 conditional novelty 6.0

    ECHO stores each completed agent turn as a source-indexed memory, selects memories for bounded context, and uses those same indices to route positive RL credit, improving long-horizon tool-use accuracy.

  5. ECHO: Prune To Act, Trace To Learn With Selective Turn Memory In Agentic RL

    cs.LG 2026-06 conditional novelty 6.0

    ECHO stores each agent turn as a source-indexed memory, reconstructs bounded contexts by selecting useful records, and routes RL credit through the same selection trace — reaching 43.4% on BrowseComp-Plus vs 28.9% (GR...

  6. ECHO: Prune To Act, Trace To Learn With Selective Turn Memory In Agentic RL

    cs.LG 2026-06 conditional novelty 6.0

    ECHO's source-indexed memory records let context reconstruction and RL credit assignment share one trace, reaching 43.4% on BrowseComp-Plus vs 28.9% (GRPO) and 36.1% (SUPO).

  7. From Time Series Analysis to Question Answering: A Survey in the LLM Era

    cs.LG 2025-06 accept novelty 6.0

    A survey proposing a taxonomy of Injective, Bridging, and Internal Alignment paradigms to evolve TSA into user-driven Time Series Question Answering with LLMs.