Copertina

How Instruction and Reasoning Data shape Post-Training: Data Quality through the Lens of Layer-wise Gradients

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Li, Ming, Li, Yanhong, Li, Ziyue, Zhou, Tianyi
Natura:	Preprint
Pubblicazione:	2025
Soggetti:	Machine Learning Artificial Intelligence Computation and Language
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

What Happened in LLMs Layers when Trained for Fast vs. Slow Thinking: A Gradient Perspective
di: Li, Ming, et al.
Pubblicazione: (2024)

Selective Reflection-Tuning: Student-Selected Data Recycling for LLM Instruction-Tuning
di: Li, Ming, et al.
Pubblicazione: (2024)

RuleR: Improving LLM Controllability by Rule-based Data Recycling
di: Li, Ming, et al.
Pubblicazione: (2024)

Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens
di: Zhao, Chengshuai, et al.
Pubblicazione: (2025)

Missing Premise exacerbates Overthinking: Are Reasoning Models losing Critical Thinking Skill?
di: Fan, Chenrui, et al.
Pubblicazione: (2025)

Schoenfeld's Anatomy of Mathematical Reasoning by Language Models
di: Li, Ming, et al.
Pubblicazione: (2025)

What makes Reasoning Models Different? Follow the Reasoning Leader for Efficient Decoding
di: Li, Ming, et al.
Pubblicazione: (2025)

Influential Language Data Selection via Gradient Trajectory Pursuit
di: Deng, Zhiwei, et al.
Pubblicazione: (2024)

Uncertainty-Aware Gradient Signal-to-Noise Data Selection for Instruction Tuning
di: Yuan, Zhihang, et al.
Pubblicazione: (2026)

How to Train Data-Efficient LLMs
di: Sachdeva, Noveen, et al.
Pubblicazione: (2024)

Interpreting and Controlling LLM Reasoning through Integrated Policy Gradient
di: Li, Changming, et al.
Pubblicazione: (2026)

BC Protocol: Structured Dual-Expert Dialogue for Eliciting High-Quality Chain-of-Thought Post-Training Data
di: Zou, Bo, et al.
Pubblicazione: (2026)

Multiplex Thinking: Reasoning via Token-wise Branch-and-Merge
di: Tang, Yao, et al.
Pubblicazione: (2026)

GraphDancer: Training LLMs to Explore and Reason over Graphs via Two-Stage Curriculum Post-Training
di: Bai, Yuyang, et al.
Pubblicazione: (2026)

TAGCOS: Task-agnostic Gradient Clustered Coreset Selection for Instruction Tuning Data
di: Zhang, Jipeng, et al.
Pubblicazione: (2024)

Reinforcement Learning on Pre-Training Data
di: Li, Siheng, et al.
Pubblicazione: (2025)

Reframing Data Value for Large Language Models Through the Lens of Plausibility
di: Rammal, Mohamad Rida, et al.
Pubblicazione: (2024)

Lean-ing on Quality: How High-Quality Data Beats Diverse Multilingual Data in AutoFormalization
di: Chan, Willy, et al.
Pubblicazione: (2025)

Can LLMs Speak For Diverse People? Tuning LLMs via Debate to Generate Controllable Controversial Statements
di: Li, Ming, et al.
Pubblicazione: (2024)

Less Noise, More Voice: Reinforcement Learning for Reasoning via Instruction Purification
di: Guo, Yiju, et al.
Pubblicazione: (2026)

Learning to Reason Efficiently with A* Post-Training
di: Opedal, Andreas, et al.
Pubblicazione: (2026)

KDRL: Post-Training Reasoning LLMs via Unified Knowledge Distillation and Reinforcement Learning
di: Xu, Hongling, et al.
Pubblicazione: (2025)

Programming Every Example: Lifting Pre-training Data Quality Like Experts at Scale
di: Zhou, Fan, et al.
Pubblicazione: (2024)

Data Efficacy for Language Model Training
di: Dai, Yalun, et al.
Pubblicazione: (2025)

ReAct Meets ActRe: When Language Agents Enjoy Training Data Autonomy
di: Yang, Zonghan, et al.
Pubblicazione: (2024)

ColorBench: Can VLMs See and Understand the Colorful World? A Comprehensive Benchmark for Color Perception, Reasoning, and Robustness
di: Liang, Yijun, et al.
Pubblicazione: (2025)

MixtureVitae: Open Web-Scale Pretraining Dataset With High Quality Instruction and Reasoning Data Built from Permissive-First Text Sources
di: Nguyen, Huu, et al.
Pubblicazione: (2025)

How Post-Training Reshapes LLMs: A Mechanistic View on Knowledge, Truthfulness, Refusal, and Confidence
di: Du, Hongzhe, et al.
Pubblicazione: (2025)

Prismatic Synthesis: Gradient-based Data Diversification Boosts Generalization in LLM Reasoning
di: Jung, Jaehun, et al.
Pubblicazione: (2025)

Can Post-Training Transform LLMs into Causal Reasoners?
di: Chen, Junqi, et al.
Pubblicazione: (2026)

Are Transformers Able to Reason by Connecting Separated Knowledge in Training Data?
di: Yin, Yutong, et al.
Pubblicazione: (2025)

Stepwise Guided Policy Optimization: Coloring your Incorrect Reasoning in GRPO
di: Chen, Peter, et al.
Pubblicazione: (2025)

How to Synthesize Text Data without Model Collapse?
di: Zhu, Xuekai, et al.
Pubblicazione: (2024)

Lil: Less is Less When Applying Post-Training Sparse-Attention Algorithms in Long-Decode Stage
di: Hu, Junhao, et al.
Pubblicazione: (2026)

How Do Your Code LLMs Perform? Empowering Code Instruction Tuning with High-Quality Data
di: Wang, Yejie, et al.
Pubblicazione: (2024)

Montessori-Instruct: Generate Influential Training Data Tailored for Student Learning
di: Li, Xiaochuan, et al.
Pubblicazione: (2024)

Superminds Test: Actively Evaluating Collective Intelligence of Agent Society via Probing Agents
di: Li, Xirui, et al.
Pubblicazione: (2026)

Understanding the Thinking Process of Reasoning Models: A Perspective from Schoenfeld's Episode Theory
di: Li, Ming, et al.
Pubblicazione: (2025)

Emergent and Subliminal Misalignment Through the Lens of Data-Mediated Transfer
di: Askin, Baris, et al.
Pubblicazione: (2026)

Is Training Data Quality or Quantity More Impactful to Small Language Model Performance?
di: Sajith, Aryan, et al.
Pubblicazione: (2024)