1
The TomeVault Instruction Corpus (2026-07)
大规模指令语料库发布，含近23万文件，可自由用于模型训练与评测。
The TomeVault Instruction Corpus De-identified structural measurements of AI instruction files (CLAUDE.md, AGENTS.md, SKILL.md, .cursorrules and relat…
2
STAIF: A Stage-wise Optimization for Complex Instruction Following
LLM复杂指令跟随难题新解法，STAIF分阶段优化显著提升模型性能
arXiv:2607.22649v1 Announce Type: cross Abstract: Following complex instructions with multiple explicit constraints remains a fundamental challenge fo…
3
FineInstructions: Scaling Synthetic Instructions to Pre-Training Scale
通过合成指令数据扩展预训练规模，突破传统监督训练数据瓶颈的新方法
arXiv:2601.22146v2 Announce Type: replace-cross Abstract: Due to limited supervised training data, large language models (LLMs) are typically pre-trai…
4
Parameter Efficient Multimodal Instruction Tuning for Romanian Vision Language Models
面向低资源语言罗马尼亚语的多模态指令微调，用参数高效方法实现视觉语言模型适配，填补非英语VLM研究空白。
arXiv:2512.14926v2 Announce Type: replace-cross Abstract: Focusing on low-resource languages is an essential step toward democratizing generative AI. …
5
Online Data Selection for Instruction Tuning via Gaussian Processes
用高斯过程在线筛选指令微调数据，让大模型训练从拼数量转向拼质量，值得一读。
arXiv:2606.30077v1 Announce Type: new Abstract: With Large Language Model (LLM) pre-training and fine-tuning shifting its focus from data volume to da…
6
VisNec: Measuring and Leveraging Visual Necessity for Multimodal Instruction Tuning
全新指标VisNec量化视觉信息必要性，精准筛选多模态训练数据，让指令微调更高效。
arXiv:2603.01195v2 Announce Type: replace-cross Abstract: The effectiveness of multimodal instruction tuning depends not only on dataset scale, but cr…
7
Two Wrongs, No Right: Auditing Social-Desirability Bias in LLM Annotators for Computational Social Science
审计LLM标注者的社会期望偏差，揭示对齐误差如何扭曲计算社会科学结论。
arXiv:2606.12426v1 Announce Type: cross Abstract: LLM annotators are increasingly used in computational social science (CSS), but it is unclear whethe…
8
A Controlled Study of Decoding-Time Truthfulness Methods on Instruction-Tuned LLMs
系统对比解码时真实性方法在指令微调LLM上的效果，揭示不同策略的优劣与适用场景
arXiv:2606.12160v1 Announce Type: new Abstract: In this work, we introduce CHAIR (Classifier of Hallucination As ImproveR), a supervised framework for…
9
Does the Question Really Matter? Training-Free Data Selection for Vision-Language SFT
发现多数视觉指令样本可通过语言模式解决，提出无需训练的数据选择方法提升跨模态学习
arXiv:2603.09715v2 Announce Type: replace Abstract: Visual instruction tuning is crucial for improving vision-language large models (VLLMs). However, …
10
Instruction Finetuning DeepSeek-R1-8B Model Using LoRA and NEFTune
探索用LoRA和NEFTune方法高效微调DeepSeek-R1-8B，降低资源消耗同时提升性能。
arXiv:2606.10392v1 Announce Type: new Abstract: Financial named-entity recognition (NER) is essential for translating unstructured financial reports a…
11
Large Language Models Are Overconfident in Their Own Responses
新研究表明指令微调大模型在自评时过度自信，揭示聊天模板对校准的关键影响。
arXiv:2606.03437v1 Announce Type: new Abstract: Prior work has shown that instruction-tuned large language models (LLMs) are less well calibrated than…
12
On the Sensitivity of Instruction-tuned LLMs to Harmful Sentences in Long Inputs
最新研究揭示指令微调LLM在长上下文输入中，对有害句子的敏感性存在显著风险。
arXiv:2510.05864v2 Announce Type: replace Abstract: Large language models (LLMs) increasingly operate on long inputs, yet their behavior when harmful …
13
Fine-Tuning Causal LLMs for Text Classification: Embedding-Based vs. Instruction-Based Approaches
对比嵌入头与指令微调两种策略，揭示资源受限下因果LLM文本分类的高效路径。
arXiv:2512.12677v2 Announce Type: replace-cross Abstract: We explore efficient strategies to fine-tune decoder-only Large Language Models (LLMs) for d…
14
PoisonForge: Task-Level Targeted Poisoning Benchmark for Instruction-Tuned LLMs
指令微调LLM面临的任务级定向投毒威胁，首个系统性基准PoisonForge发布，助力模型安全评估。
arXiv:2605.23168v1 Announce Type: cross Abstract: When practitioners fine-tune LLMs on unvetted datasets, an adversary can exploit the data supply cha…
15
Task-conditioned probing of instruction-tuned multimodal LLMs: Region-specific brain alignment patterns under naturalistic stimuli
探索指令微调多模态大模型在自然刺激下的脑区对齐模式，交叉验证AI与神经科学
arXiv:2506.08277v3 Announce Type: replace-cross Abstract: Recent voxel-wise multimodal brain encoding studies have shown that multimodal large languag…
16
StrLoRA: Towards Streaming Continual Visual Instruction Tuning for MLLMs
StrLoRA提出流式持续视觉指令微调新方法，有效缓解多模态大模型在序列任务中的灾难性遗忘。
arXiv:2605.16353v1 Announce Type: new Abstract: Continual Visual Instruction Tuning (CVIT) enables Multimodal Large Language Models to incrementally a…