跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+1111新增论文
+874其中 AI 相关
+133已有论文修订
+97新增 AI 导读

arXiv 最近完成扫描:2026-11-20 08:48 · 全站最近入库:2026-11-19 19:09各信源进度 →

统计口径与数据新鲜度

统计 2026-11-19 09:19 — 2026-11-20 09:19(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-20 08:48最近发现新增版本:2026-11-19 19:09库内论文最新更新时间:2026-11-19 02:59

每分钟刷新已保存的数据。空检查不代表采集异常;新收录可能包含历史论文。AI 导读按新增版本计数。

27,982 已收录论文23,249 AI 相关8,559 有修订112 HF 社区发现

浏览论文库

时间范围跟随排序:版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常,查看状态 →

找到 3 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-20 08:46(北京时间)
版本更新 · v2arXiv:2609.02083

XMerge: Cross-Axis Selection and Reconstructive Layer Merging for LLM Depth Compression

Jundong Hu, Shekar Ramachandran

原文摘要

Removing complete transformer layers preserves a standard serving architecture, but existing depth-compression methods can lose substantial quality, and the loss varies unpredictably across models. We introduce XMerge, a post-training method with two components. Cross-axis selection identifies a block with low relative-magnitude and angular hidden-state change, and local boundary reconstruction re-fits the adjacent surviving block to match the original two-block output. XMerge uses no task labels or end-to-end fine-tuning, and it introduces neither architectural changes nor additional inference-time parameters. Across seven Llama and Qwen backbones (0.5B-8B), five published baselines, and three layer-reduction levels, its advantage over baselines is largest at the most aggressive removal: at k=4 it ranks first on six of seven backbones on CORE (a 22-task aggregate) and, separately, on six of seven on MMLU (five of seven on both at once), while avoiding the large perplexity increases of several competing operators. In a task-level bootstrap, the 95% confidence intervals for the three largest CORE margins exclude zero; the remaining margins are consistent with ties. Across the 14 (model, regime) cells it is also the only evaluated operator that never collapses, ranking top-2 in both zero-shot and in-context regimes; on a first calibration probe (one backbone) it is the best-calibrated operator. Ablations show that local reconstruction provides most of the gain, while cross-axis fusion helps when the two selection axes disagree. The additional construction cost is recovered through per-token decode savings after roughly tens of thousands of requests.

cs.LG · cs.CL

版本更新 · v2arXiv:2609.01852

The Memory Trust Gap: Reliance on Stale Evidence Across Model Scales in LLM Agents

Jundong Hu, Shekar Ramachandran

原文摘要

Persistent-memory agents can follow stale information despite current authoritative evidence. We study stale-evidence reliance across Qwen3 0.6/1.7/4/8B checkpoints using two controlled, action-scored suites: a Benefit suite requiring stored information and a Safety suite providing current evidence as static prompt text. We distinguish stale-value reliance from net accuracy loss relative to no memory. Benefit-suite reliance is 0.92-1.00 across checkpoints. In the Safety suite, larger checkpoints resist a plain stale note but incur substantial harm when it appears newer. A full $2\times2\times2\times2$ factorial manipulates label, recency, source authority, and position. Direct cross-size contrasts identify stronger recency effects in the larger tested checkpoints and a position-effect sign reversal within Qwen3. Metadata improves accuracy at all four checkpoints, with larger gains at 4B/8B; pre-resolution yields near-ceiling accuracy across all four. Llama-Instruct comparisons provide cross-family evidence for selected effects, while RGB and MisBench corroborate stale-value reliance. A framing control shows that the failure extends beyond memory-labeled content. These results characterize cue-dependent vulnerability across tested checkpoints without isolating capability or evaluating a complete memory lifecycle.

cs.AI · cs.CL

版本更新 · v2arXiv:2609.01587

The Structure of Quantization Damage in LLMs: Why the Next Bit Should Be Spent Globally

Jundong Hu, Shekar Ramachandran

原文摘要

Post-training quantization (PTQ) is widely used to reduce the cost of serving large language models (LLMs), but its accuracy cost is uneven and is often tuned per model. We study where quantization damage occurs and how to allocate a small additional precision budget. Using causal mixed-precision intervention as ground truth (raise each layer to 8-bit in turn and measure the accuracy it recovers) across 9 open-weight models in 4 architecture families, we test 3 intuitive hypotheses: that quantization damage lives in task circuits, where the model computes, or in weight statistics. None of them predicts which layers benefit from restored precision. Recovery is instead diffuse: for 8 of 9 models, recovering 75% of the gap takes roughly half the layers; the lone exception, Qwen3-8B, is sharply concentrated. At a matched precision budget, spending it globally on finer quantization granularity beats locally repairing the most recoverable layers for all 8 group-128-compatible models (all but OpenLLaMA, whose width rules out group-128), by 21-52 points, including the concentrated Qwen3-8B. We report 2 secondary findings: the residual is budget-limited (8-bit is near-lossless in our evaluation across RTN, GPTQ, and AWQ), and the location of peak recovery correlates with architecture within a family, though not across families. Within this budget setting, global granularity is a better default than selectively protecting critical layers. More broadly, cheap signals that correlate with quantization damage do not necessarily identify where restoring precision improves accuracy; this must be tested with causal intervention.

cs.LG · cs.CL