Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2609.08966

Good Pretraining, Bad SFT: Checkpoint Quality Across the Training Stack

Sohir Maskey·Philipp Scholl·Jonas Knupp·Pit Neitemeier·Sascha Wirges
View original source ↗Read source document ↗Export metadata ↓

Abstract

Language-model checkpoints are commonly selected by pretraining loss or benchmark scores, assuming that the highest-scoring checkpoint will remain the best starting point for subsequent training. We show that this assumption can fail in a full 30B mixture-of-experts training pipeline. The checkpoints that perform better after the full downstream training stack also have higher solution density, i.e., retain downstream performance under local weight perturbations.

Explore related subjects

cs.AIcs.CL

Keep this discovery

BibTeXRIS

Sohir Maskey, Philipp Scholl, Jonas Knupp, Pit Neitemeier, Sascha Wirges. 2026-09-08. Good Pretraining, Bad SFT: Checkpoint Quality Across the Training Stack. https://arxiv.org/abs/2609.08966

Cite the original work for its findings. Save a collection to share your selection of sources.

Discover connections

cs.AI subjectcs.CL subject

Connections use source metadata and explicit phrase matches, not verified experimental comparisons.

KEEP EXPLORING

Related papers

arXiv2026

CyrillicQA: The Influence of Phonetically Encoded Secret Language on LLM Performance

Due to the selection of their training data, large language models (LLMs) perform best on standard-language inputs from languages using the Latin alphabet with large speaker populations, while disadvantaging other language varieties. Nevertheless, they can also be a versatile tool for preserving precisely such endangered languages. But do they also possess the necessary creativity and capacity for abstraction to decode phonetically encoded language the same way humans do?

cs.CL↗
arXiv2026

When Robots Mishear Us: Mapping the Safety Risks of Voice-Controlled Embodied AI

We investigate whether automatic speech recognition (ASR) errors in user input can lead to unsafe outputs from Embodied AI (EAI) models. We find that ASR errors can lead to harmful instructions being accepted and executed by EAI models, thereby reducing safety. We simulate ASR errors and combine them with existing safety benchmarks (SafeAgentBench and POEX) to evaluate how different errors affect embodied AI safety. We find that some of them preserve semantic structure but increase harmful ambiguity, while others weaken the model refusal behaviour and allow unsafe plans to be generated and executed. We show that in some cases automatic correction of ASR errors can reduce the risk, but this is not always effective. Overall, we show that ASR errors lead to significant safety risks for embodied AI.

cs.AI↗
arXiv2026

AI Alignment through a Game-theoretic Lens: A Survey

As large language models and increasingly capable AI agents are deployed in high-risk settings, aligning them with complex human values has become a central challenge. Existing alignment methods, while effective in improving helpfulness, harmlessness, and controllability, often struggle to capture real-world preferences that are context-dependent, non-transitive, and shaped by dynamic multi-party interactions. This survey reviews AI alignment through a game-theoretic lens. Specifically, it organizes recent progress around key game-theoretic elements and synthesizes the literature along three challenges: preference diversity, alignment priority, and temporal dynamics. This perspective clarifies where current alignment methods genuinely benefit from game-theoretic analysis, where the framework is looser, and what challenges remain in building robust, adaptive, and verifiable AI systems.

cs.AI↗