mlsanityData to 14 Oct 2026

LLM Post-Training as Brownfield Maintenance: An Industrial Perspective on Dataware Engineering

arXiv 2608.31102 ↗First posted 31 Aug 2026cs.SE cs.AI cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Industrial post-training is a brownfield regime. Teams inherit a deployed checkpoint and must land targeted improvements under fixed compute and mixture budgets without regressing the rest. The maintained artifact is increasingly dataware: behavior governed by a curated post-training mixture, updated via bounded mixture patches rather than clean-slate retraining. From an industrial code-generation improvement effort, we offer a maintainer's perspective on why this work is hard in practice, distilling three recurring challenges, zero-sum mixture design, yield as the binding metric, and end-to-end integration under uncertainty, and arguing that progress depends less on one-off recipes than on an engineering discipline for programming dataware. In our case study, interventions that raised the conversion of teacher distillation into usable training data increased accepted supervision by 2.84 times while using the same solution teacher and four solution attempts per candidate problem. In our primary evaluation, the yield-engineered patch improved CodeForces pass@1 by +2.59 points (+3.11 pass@3) and held-out LiveCodeBench v6 pass@1 by +6.11 (+8.05 pass@3), all statistically significant across 16 stochastic evaluations of each benchmark from one fixed checkpoint per condition, with internal AIME and MATH regression suites within tolerance.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 15 of its 26 references that are in the map. First posted on arXiv 31 Aug 2026.

Builds on

15 of 26 references.

References · 15

  1. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  2. 2022Self-Instruct: Aligning Language Models with Self-Generated Instructionscited by 2,724
  3. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 2,004
  4. 2023DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretrainingcited by 399
  5. 2024RegMix: Data Mixture as Regression for Language Model Pre-trainingcited by 177
  6. 2024Adding Error Bars to Evals: A Statistical Approach to Language Model Evaluationscited by 148
  7. 2025OpenCodeReasoning: Advancing Data Distillation for Competitive Codingcited by 113
  8. 2023DoGE: Domain Reweighting with Generalization Estimationcited by 90
  9. 2025Effectively Controlling Reasoning Models through Thinking Interventioncited by 65
  10. 2024InsCL: A Data-efficient Continual Learning Paradigm for Fine-tuning Large Language Models with Instructionscited by 58
  11. 2025SwS: Self-aware Weakness-driven Problem Synthesis in Reinforcement Learning for LLM Reasoningcited by 42
  12. 2024AutoDetect: Towards a Unified Framework for Automated Weakness Detection in Large Language Modelscited by 9
3 more
  1. 2025Dynamic Stability of LLM-Generated Codecited by 4
  2. 2026Logics-STEM: Empowering LLM Reasoning via Failure-Driven Post-Training and Document Knowledge Enhancementcited by 4
  3. 2025The Valley of Code Reasoning: Scaling Knowledge Distillation of Large Language Modelscited by 2

Built on by

Not cited yet.