# Unveiling the Basin-Like Loss Landscape in Large Language Models

arXiv: https://arxiv.org/abs/2505.17646 · HTML: https://mlsanity.com/paper/2505.17646/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- Builds on 82 of its 102 references that are in the map.
- First posted on arXiv 23 May 2025.

## Builds on (most cited first, 30 of 82)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [Towards Deep Learning Models Resistant to Adversarial Attacks](https://mlsanity.com/paper/1706.06083/) (2017)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [Neural Machine Translation of Rare Words with Subword Units](https://mlsanity.com/paper/1508.07909/) (2015)
- [Neural Tangent Kernel: Convergence and Generalization in Neural Networks](https://mlsanity.com/paper/1806.07572/) (2018)
- [InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks](https://mlsanity.com/paper/2312.14238/) (2023)
- [MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI](https://mlsanity.com/paper/2311.16502/) (2023)
- [Jailbroken: How Does LLM Safety Training Fail?](https://mlsanity.com/paper/2307.02483/) (2023)
- [ChartQA: A Benchmark for Question Answering about Charts with Visual and Logical Reasoning](https://mlsanity.com/paper/2203.10244/) (2022)
- [Jailbreaking Black Box Large Language Models in Twenty Queries](https://mlsanity.com/paper/2310.08419/) (2023)
- [Sharpness-Aware Minimization for Efficiently Improving Generalization](https://mlsanity.com/paper/2010.01412/) (2020)
- [On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima](https://mlsanity.com/paper/1609.04836/) (2016)
- [Certified Adversarial Robustness via Randomized Smoothing](https://mlsanity.com/paper/1902.02918/) (2019)
- [Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!](https://mlsanity.com/paper/2310.03693/) (2023)
- [Visualizing the Loss Landscape of Neural Nets](https://mlsanity.com/paper/1712.09913/) (2017)
- [Red Teaming Language Models with Language Models](https://mlsanity.com/paper/2202.03286/) (2022)
- [Reconciling modern machine learning practice and the bias-variance trade-off](https://mlsanity.com/paper/1812.11118/) (2018)
- [A Simple and Effective Pruning Approach for Large Language Models](https://mlsanity.com/paper/2306.11695/) (2023)
- [A Convergence Theory for Deep Learning via Over-Parameterization](https://mlsanity.com/paper/1811.03962/) (2018)
- [Aligning AI With Shared Human Values](https://mlsanity.com/paper/2008.02275/) (2020)
- [Safe RLHF: Safe Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2310.12773/) (2023)
- [Linear Mode Connectivity and the Lottery Ticket Hypothesis](https://mlsanity.com/paper/1912.05671/) (2019)
- [Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNs](https://mlsanity.com/paper/1802.10026/) (2018)

## Built on by (most cited first, 13 of 13)

- [Preconditioned Norms: A Unified Framework for Steepest Descent, Quasi-Newton and Adaptive Methods](https://mlsanity.com/paper/2510.10777/) (2025)
- [RCP-Merging: Merging Long Chain-of-Thought Models with Domain-Specific Models by Considering Reasoning Capability as Prior](https://mlsanity.com/paper/2508.03140/) (2025)
- [Resource-Efficient Reinforcement for Reasoning Large Language Models via Dynamic One-Shot Policy Refinement](https://mlsanity.com/paper/2602.00815/) (2026)
- [LC-QAT: Data-Efficient 2-Bit QAT for LLMs via Linear-Constrained Vector Quantization](https://mlsanity.com/paper/2606.10531/) (2026)
- [Defending Unauthorized Model Merging via Dual-Stage Weight Protection](https://mlsanity.com/paper/2511.11851/) (2025)
- [ReGA: Model-Based Safeguard for LLMs via Representation-Guided Abstraction](https://mlsanity.com/paper/2506.01770/) (2025)
- [Unveiling Multi-regime Patterns in SciML: Distinct Failure Modes and Regime-specific Optimization](https://mlsanity.com/paper/2605.29153/) (2026)
- [Understanding Quantization-Aware Training: Gradients at Quantized Weights Bias to the Low-Loss Basin](https://mlsanity.com/paper/2606.09012/) (2026)
- [You Snooze, You Lose: Automatic Safety Alignment Restoration through Neural Weight Translation](https://mlsanity.com/paper/2605.04992/) (2026)
- [Beyond Perplexity: A Geometric and Spectral Study of Low-Rank Pre-Training](https://mlsanity.com/paper/2605.13652/) (2026)
- [A Gravitational Interpretation of Safety Reversion under Fine-Tuning](https://mlsanity.com/paper/2606.28525/) (2026)
- [The Blessing of Dimensionality: How Near-Orthogonality in High-Dimensional Spaces Explains Temporal Portability](https://mlsanity.com/paper/2607.20301/) (2026)
- [Beyond Quadratic Loss: The Stability Phase Diagram of Adam](https://mlsanity.com/paper/2609.18314/) (2026)

## Abstract (arXiv)

We discover the emergence of textitbasins in the loss landscape of large language models. As model scale increases, LLMs become progressively more resilient to random perturbations in the parameter space, giving rise to expansive stability regions where models exhibit nearly identical performance, but outside of which their capabilities collapse. We observe that pre-training creates a textitbasic capability basin, and subsequent alignment fine-tuning forms textitspecific capability basins (e.g., safety, math, coding). Thus, we argue that benign fine-tuning confined to the basin should preserve prior capabilities. Besides, we also analyze the loss landscape for worst-case directions, which is consistently sharp and detrimental. We find that adversarial fine-tuning moves along the nearly worst-case directions, thus rapidly degrading model capabilities. Finally, we provide a theoretical analysis demonstrating that the basin size bounds the performance degradation of any fine-tuning, including the adversarial ones, while also guaranteeing the model robustness w.r.t. input perturbations, suggesting the benefit of enlarging basins.