Separating Constraint Compliance from Semantic Accuracy: A Novel Benchmark for Evaluating Instruction-Following Under Compression

• URL: http://arxiv.org/abs/2512.17920v1
• Date: Tue, 02 Dec 2025 13:25:48 GMT
• Title: Separating Constraint Compliance from Semantic Accuracy: A Novel Benchmark for Evaluating Instruction-Following Under Compression
• Abstract summary: Large language models (LLMs) exhibit degraded performance under prompt compression.<n>We introduce the CompressionDecay Test (CDCT), a benchmark that measures constraint compliance (CC) and semantic accuracy (SA)<n>We observe a universal U-curve pattern in constraint compliance (97.2% prevalence), with violations peaking at medium compression.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models (LLMs) exhibit degraded performance under prompt compression, but the mechanisms remain poorly understood. We introduce the Compression-Decay Comprehension Test (CDCT), a benchmark that independently measures constraint compliance (CC) and semantic accuracy (SA) across compression levels. We evaluate 9 frontier LLMs across 8 concepts using 5 compression levels from extreme (c=0.0, ~2 words) to none (c=1.0, ~135 words). A three-judge LLM jury achieves almost perfect inter-rater agreement on CC (Fleiss' \k{appa}=0.90). We observe a universal U-curve pattern in constraint compliance (97.2% prevalence), with violations peaking at medium compression (c=0.5, ~27 words). Counterintuitively, models perform better at extreme compression than medium lengths. The dimensions are statistically orthogonal (r=0.193, p=0.084), with constraint effects 2.9x larger than semantic effects. Experimental validation via RLHF ablation confirms our constraint salience hypothesis: removing "helpfulness" signals improves CC by 598% on average (71/72 trials, p<0.001), with 79% achieving perfect compliance. This demonstrates that RLHF-trained helpfulness behaviors are the dominant cause of constraint violations at medium compression. Reasoning models outperform efficient models by 27.5% (Cohen's d=0.96). Our findings reveal a fundamental tension between RLHF alignment and instruction-following, providing actionable guidelines for improving deployed systems.

Related papers

• On the Interaction Between Model Compression and Test-Time Adaptation [16.18375474723875]
  We analyze how structured compression affects a model's ability to adapt under distribution shift.<n>We introduce a diagnostic framework that examines representational expressivity and adaptation subspace compatibility.
  arXiv  Detail & Related papers  (2026-09-03T09:49:29Z)
• CHERRY: Compressed Hierarchical Experts with Recurrent Representational Yield [1.2203679037551938]
  We study three techniques for training compute-efficient language models.<n>SGT concentrates supervision on the 15% of output tokens that carry semantic payload.<n> depth compression with recurrent recovery.<n>We validate these techniques on CHERRY-1.8B, a Korean foundation model.
  arXiv  Detail & Related papers  (2026-06-30T15:14:38Z)
• CaliDist: Calibrating Large Language Models via Behavioral Robustness to Distraction [51.56484100374058]
  Existing calibration methods for Large Language Models (LLMs) often overlook a critical dimension of trustworthiness: a model's behavioral robustness.<n>We introduce textscCaliDist, a novel post-hoc calibration approach that directly measures and penalizes a model's susceptibility to distraction.<n>textscCaliDist consistently achieves lower Expected Error (ECE) and Brier Score compared with strong baselines.
  arXiv  Detail & Related papers  (2026-06-04T07:27:53Z)
• Does Compression Preserve Uncertainty? A Unified Benchmark for Quantized and Sparse LLMs via Conformal Prediction [21.411353047208777]
  We benchmark 12 large language models (LLMs) under various compression configurations across five NLP tasks.<n>Our experiments reveal that: (I) compression frequently decouples accuracy from uncertainty; (II) larger models absorb compression-induced uncertainty far more effectively than smaller ones; and (III) uncertainty inflation is often threshold-like rather than gradual.
  arXiv  Detail & Related papers  (2026-06-01T08:00:17Z)
• AGORA: Adapter-Grounded Observation-Action Retention for Inference-Free Prompt Compression in LLM Agents [3.9004035576487817]
  We show that token-level extractive compressors are structurally inappropriate for LLM agents.<n>We introduce AGORA, an inference-free step-level compressor combining a structural prompt, an always-keep floor for format-critical and recency-critical content.
  arXiv  Detail & Related papers  (2026-05-26T06:29:44Z)
• Amplifying, Not Learning: Fine-Tuned AI Text Detectors Amplify a Pretrained Direction [51.56484100374058]
  Text detectors amplify a pretrained typicality axis; they do not construct an AI-vs-human boundary.<n>On raw encoders before any task supervision, projecting onto centroid(AI)-centroid(HC3) achieves NYT-vs-HC3 AUROC 0.806/0.944/0.834 across three architectures.<n>On RoBERTa-base, raw projection exceeds fine-tuning; on RoBERTa-base, full fine-tuning reduces discrimination below raw on both fluent-formal populations tested.
  arXiv  Detail & Related papers  (2026-05-20T19:08:38Z)
• Models Recall What They Violate: Constraint Adherence in Multi-Turn LLM Ideation [0.0]
  We introduce DriftBench, a benchmark for evaluating constraint adherence in scientific ideation.<n>We find that iterative pressure reliably increases structural complexity and often reduces adherence to original constraints.<n>We release all briefs, prompts, rubrics, transcripts, and scores as an open benchmark.
  arXiv  Detail & Related papers  (2026-04-30T15:46:33Z)
• BadCLIP++: Stealthy and Persistent Backdoors in Multimodal Contrastive Learning [73.46118996284888]
  Research on backdoor attacks against multimodal contrastive learning models faces two key challenges: stealthiness and persistence.<n>We propose BadCLIP++, a unified framework that tackles both challenges.<n>For stealthiness, we introduce a semantic-fusion QR micro-trigger that embeds imperceptible patterns near task-relevant regions.<n>For persistence, we stabilize trigger embeddings via radius shrinkage and centroid alignment.
  arXiv  Detail & Related papers  (2026-02-19T08:31:16Z)
• Information Fidelity in Tool-Using LLM Agents: A Martingale Analysis of the Model Context Protocol [69.11739400975445]
  We introduce the first theoretical framework for analyzing error accumulation in Model Context Protocol (MCP) agents.<n>We show that cumulative distortion exhibits linear growth and high-probability deviations bounded by $O(sqrtT)$.<n>Key findings include: semantic weighting reduces distortion by 80%, and periodic re-grounding approximately every 9 steps suffices for error control.
  arXiv  Detail & Related papers  (2026-02-10T21:08:53Z)
• Towards Efficient Large Language Reasoning Models via Extreme-Ratio Chain-of-Thought Compression [55.63153956934198]
  Chain-of-Thought (CoT) reasoning successfully enhances the reasoning capabilities of Large Language Models (LLMs)<n>Existing CoT compression methods often suffer from a critical loss of logical fidelity at high compression ratios.<n>We propose a novel EXTreme-RAtio Chain-of-Thought Compression framework, termed Extra-CoT, which aggressively reduces the token budget while preserving answer accuracy.
  arXiv  Detail & Related papers  (2026-02-09T06:57:15Z)
• Latent Sculpting for Zero-Shot Generalization: A Manifold Learning Approach to Out-of-Distribution Anomaly Detection [2.8547732086436306]
  A fundamental limitation of supervised deep learning is "Generalization Collapse"<n>We propose Latent Sculpting, a hierarchical two-stage representation learning framework.<n>We report an 88.89% detection rate on "Infiltration" scenarios.
  arXiv  Detail & Related papers  (2025-12-19T11:37:02Z)
• Detecting AI Hallucinations in Finance: An Information-Theoretic Method Cuts Hallucination Rate by 92% [4.693270291878929]
  Large language models (LLMs) produce fluent but unsupported answers - hallucinations.<n>We propose ECLIPSE, a framework that treats hallucination as a mismatch between a model's semantic entropy and the capacity of available evidence.
  arXiv  Detail & Related papers  (2025-12-02T05:25:48Z)
• Choose Your Model Size: Any Compression of Large Language Models Without Re-Computation [10.376875638696504]
  This work presents Any Compression via Iterative Pruning (ACIP), a novel algorithmic approach to determine a compression-performance trade-off.<n>We use an SVD-reparametrization of linear layers and iteratively prune their singular values with a sparsity-inducing penalty.<n>We show that ACIP seamlessly complements common quantization-based compression techniques.
  arXiv  Detail & Related papers  (2025-02-03T18:40:58Z)
• UNComp: Can Matrix Entropy Uncover Sparsity? -- A Compressor Design from an Uncertainty-Aware Perspective [85.08718140718707]
  UNComp is an uncertainty-aware framework that uncovers sparsity patterns that can be used for adaptive compression.<n>By focusing on uncertainty to analyze the sparsity pattern in detail, UNComp reduces the KV cache size to 4.74% of the original, achieves a 6% prefill speedup, and improves throughput by 6.4x.
  arXiv  Detail & Related papers  (2024-10-04T02:32:36Z)
• FDC: Fast KV Dimensionality Compression for Efficient LLM Inference [11.194752361478567]
  FDC is a fast KV dimensionality compression system that eliminates the decompression overhead incurred in the existing KV dimensionality compression system, Palu, and reduces attention time.<n>In experiments, FDC can reduce Job Completion Time (JCT) by up to 64%, and delivers up to 1.97X throughput under the same latency.<n>When state-of-the-art eviction and quantization methods are combined with FDC, they exhibit similar improvements compared to those combined with Palu.
  arXiv  Detail & Related papers  (2024-08-07T22:10:26Z)
• Beyond Perplexity: Multi-dimensional Safety Evaluation of LLM Compression [33.45167213570976]
  We investigate the impact of model compression on four dimensions: (1) degeneration harm, i.e., bias and toxicity in generation; (2) representational harm, i.e., biases in discriminative tasks; (3) dialect bias; and(4) language modeling and downstream task performance. Our analysis reveals that compression can lead to unexpected consequences.
  arXiv  Detail & Related papers  (2024-07-06T05:56:22Z)
• Communication-Efficient Distributed Learning with Local Immediate Error Compensation [95.6828475028581]
  We propose the Local Immediate Error Compensated SGD (LIEC-SGD) optimization algorithm. LIEC-SGD is superior to previous works in either the convergence rate or the communication cost.
  arXiv  Detail & Related papers  (2024-02-19T05:59:09Z)
• Activations and Gradients Compression for Model-Parallel Training [85.99744701008802]
  We study how simultaneous compression of activations and gradients in model-parallel distributed training setup affects convergence. We find that gradients require milder compression rates than activations. Experiments also show that models trained with TopK perform well only when compression is also applied during inference.
  arXiv  Detail & Related papers  (2024-01-15T15:54:54Z)
• Optimal Rate Adaption in Federated Learning with Compressed Communications [28.16239232265479]
  Federated Learning incurs high communication overhead, which can be greatly alleviated by compression for model updates. tradeoff between compression and model accuracy in the networked environment remains unclear. We present a framework to maximize the final model accuracy by strategically adjusting the compression each iteration.
  arXiv  Detail & Related papers  (2021-12-13T14:26:15Z)
• On Biased Compression for Distributed Learning [55.89300593805943]
  We show for the first time that biased compressors can lead to linear convergence rates both in the single node and distributed settings. We propose several new biased compressors with promising theoretical guarantees and practical performance.
  arXiv  Detail & Related papers  (2020-02-27T19:52:24Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.