LLM Post-Training as Brownfield Maintenance: An Industrial Perspective on Dataware Engineering
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-e
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- PossiblePossibly related (embedding) · 50%From experiment to production: A reliable architecture for version-controlled MLOps →
- FuzzySimilar title/name (fuzzy) · 87%patchy631/ai-engineering-hub →
“Fuzzy title match (0.94): “LLM Post-Training as Brownfield Maintenance: An Industrial P” ≈ “patchy631/ai-engineering-hub””
- FuzzySimilar title/name (fuzzy) · 84%stas00/ml-engineering →
“Fuzzy title match (0.92): “LLM Post-Training as Brownfield Maintenance: An Industrial P” ≈ “stas00/ml-engineering””
- FuzzySimilar title/name (fuzzy) · 59%cobusgreyling/loop-engineering →
“Fuzzy title match (0.73): “LLM Post-Training as Brownfield Maintenance: An Industrial P” ≈ “cobusgreyling/loop-engineering””
- FuzzySimilar title/name (fuzzy) · 59%NirDiamant/Prompt_Engineering →
“Fuzzy title match (0.73): “LLM Post-Training as Brownfield Maintenance: An Industrial P” ≈ “NirDiamant/Prompt_Engineering””
- FuzzySimilar title/name (fuzzy) · 59%rohitg00/ai-engineering-from-scratch →
“Fuzzy title match (0.73): “LLM Post-Training as Brownfield Maintenance: An Industrial P” ≈ “rohitg00/ai-engineering-from-scratch””
- LinkedLinked via arxiv author · 85%Gopi Krishnan Rajbahadur →
“LLM Post-Training as Brownfield Maintenance: An Industrial Perspective on Dataware Engineering”
- LinkedLinked via arxiv author · 85%Amir M. Ebrahimi →
“LLM Post-Training as Brownfield Maintenance: An Industrial Perspective on Dataware Engineering”