$Cover Image$

Do Instruction-Tuned Models Always Perform Better Than Base Models? Evidence from Math and Domain-Shifted Benchmarks

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Munjal, Prateek, Christophe, Clement, Rajan, Ronnie, Kanithi, Praveenkumar
Format:	Preprint
Published:	2026
Subjects:	Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Overalignment in Frontier LLMs: An Empirical Study of Sycophantic Behaviour in Healthcare
by: Christophe, Clément, et al.
Published: (2026)

MEDIC: Comprehensive Evaluation of Leading Indicators for LLM Safety and Utility in Clinical Applications
by: Kanithi, Praveenkumar, et al.
Published: (2024)

Beyond Metrics: A Critical Analysis of the Variability in Large Language Model Evaluation Frameworks
by: Pimentel, Marco AF, et al.
Published: (2024)

Cross-Examination Framework: A Task-Agnostic Diagnostic for Information Fidelity in Text-to-Text Generation
by: Raha, Tathagata, et al.
Published: (2026)

Do Transformer World Models Give Better Policy Gradients?
by: Ma, Michel, et al.
Published: (2024)

Benchmarking GNN Models on Molecular Regression Tasks with CKA-Based Representation Analysis
by: Rajan, et al.
Published: (2026)

Bigger is not Always Better: Scaling Properties of Latent Diffusion Models
by: Mei, Kangfu, et al.
Published: (2024)

MathScale: Scaling Instruction Tuning for Mathematical Reasoning
by: Tang, Zhengyang, et al.
Published: (2024)

OpenMathInstruct-1: A 1.8 Million Math Instruction Tuning Dataset
by: Toshniwal, Shubham, et al.
Published: (2024)

From Language Modeling to Instruction Following: Understanding the Behavior Shift in LLMs after Instruction Tuning
by: Wu, Xuansheng, et al.
Published: (2023)

Benchmarking Large Language Models for Math Reasoning Tasks
by: Seßler, Kathrin, et al.
Published: (2024)

Limitations of Public Chest Radiography Datasets for Artificial Intelligence: Label Quality, Domain Shift, Bias and Evaluation Challenges
by: Rafferty, Amy, et al.
Published: (2025)

Triplets Better Than Pairs: Towards Stable and Effective Self-Play Fine-Tuning for LLMs
by: Wang, Yibo, et al.
Published: (2026)

What Makes Looped Transformers Perform Better Than Non-Recursive Ones
by: Gong, Zixuan, et al.
Published: (2025)

A Unified Graph Language Model for Multi-Domain Multi-Task Graph Alignment Instruction Tuning
by: Chen, Haibo, et al.
Published: (2026)

Is Bigger Edit Batch Size Always Better? -- An Empirical Study on Model Editing with Llama-3
by: Yoon, Junsang, et al.
Published: (2024)

Tabular Data Generation Models: An In-Depth Survey and Performance Benchmarks with Extensive Tuning
by: Kindji, G. Charbel N., et al.
Published: (2024)

Do Better Volatility Forecasts Lead to Better Portfolios? Evidence from Graph Neural Networks
by: Wade, Rylan
Published: (2026)

Two Minds Better Than One: Collaborative Reward Modeling for LLM Alignment
by: Zhang, Jiazheng, et al.
Published: (2025)

Bridging Language Barriers in Healthcare: A Study on Arabic LLMs
by: Saadi, Nada, et al.
Published: (2025)

Mortgage Language Model: Domain-Adaptive Pretraining with Residual Instruction, Alignment Tuning, and Task-Specific Routing
by: Jain, Manish, et al.
Published: (2025)

Instruction-tuned Language Models are Better Knowledge Learners
by: Jiang, Zhengbao, et al.
Published: (2024)

Are Greedy Task Orderings Better Than Random in Continual Linear Regression?
by: Tsipory, Matan, et al.
Published: (2025)

Graph Neural Networks Do Not Always Oversmooth
by: Epping, Bastian, et al.
Published: (2024)

WirelessMathBench: A Mathematical Modeling Benchmark for LLMs in Wireless Communications
by: Li, Xin, et al.
Published: (2025)

Instruction Tuning Chronologically Consistent Language Models
by: He, Songrun, et al.
Published: (2025)

Better World Models Can Lead to Better Post-Training Performance
by: Gupta, Prakhar, et al.
Published: (2025)

TuneShift-KD: Knowledge Distillation and Transfer for Fine-tuned Models
by: Guan, Yushi, et al.
Published: (2026)

Attackers Can Do Better: Over- and Understated Factors of Model Stealing Attacks
by: Oliynyk, Daryna, et al.
Published: (2025)

Is More Context Always Better? Examining LLM Reasoning Capability for Time Interval Prediction
by: Cao, Yanan, et al.
Published: (2026)

Masked Diffusion Models are Secretly Learned-Order Autoregressive Models
by: Garg, Prateek, et al.
Published: (2025)

The Vulnerability of Language Model Benchmarks: Do They Accurately Reflect True LLM Performance?
by: Banerjee, Sourav, et al.
Published: (2024)

Should We Always Train Models on Fine-Grained Classes?
by: Pirovano, Davide, et al.
Published: (2025)

Benchmarks Saturate When The Model Gets Smarter Than The Judge
by: Ballon, Marthe, et al.
Published: (2026)

Solving for X and Beyond: Can Large Language Models Solve Complex Math Problems with More-Than-Two Unknowns?
by: Kao, Kuei-Chun, et al.
Published: (2024)

Silent Commitment Failure in Instruction-Tuned Language Models: Evidence of Governability Divergence Across Architectures
by: Ruddell, Gregory M.
Published: (2026)

Towards Understanding Why FixMatch Generalizes Better Than Supervised Learning
by: Li, Jingyang, et al.
Published: (2024)

Clustering-Based Validation Splits for Model Selection under Domain Shift
by: Napoli, Andrea, et al.
Published: (2024)

Empowering Domain-Specific Language Models with Graph-Oriented Databases: A Paradigm Shift in Performance and Model Maintenance
by: Di Pasquale, Ricardo, et al.
Published: (2024)

GCAL: Adapting Graph Models to Evolving Domain Shifts
by: Qiao, Ziyue, et al.
Published: (2025)