Paper Overview
Field: ML Authors: Samuel Kushnir, Kimia Noorbakhsh, Kavya Sreedhar, Liqun Cheng, Ming Liu, Parthasarathy Ranganathan, Mohammad Alizadeh, Fred Kjolstad, Suvinay Subramanian Published: 2026-09-04 arXiv: 2609.05364
Summary
Machine-learning performance modeling is a uniquely hostile terrain for long-lived software: assumptions baked into today's abstractions are invalidated by tomorrow's models and systems, forcing perpetual refactoring of performance-modeling frameworks. Meanwhile, AI coding agents have become fast and capable enough that regenerating an entire library is cheaper than paying down the tech debt of incrementally patching it.
This paper describes SMART, a rigorous symbolic performance-modeling library for ML systems whose main branch contains almost no code:
- The repository is a DAG of self-contained natural-language design docs.
- Coding sub-agents regenerate the implementation from only the docs on new version updates.
- Every human change is a natural-language edit to a doc — self-documenting by construction.
- Symbolic (SymPy) cost expressions
- A fast analytical summary mode for large-scale sweeps
- A slow modulo-scheduling mode for fine-grained scheduling studies
Key Ideas
Two elements make regeneration reliable:
1. Design-doc style built around step-by-step worked examples, which act as in-context demonstrations for the generation agents. 2. A minimally recursively defined operational IR featuring:
Results
Regenerated implementations reproduce hand-audited reference models — including DeepSeek-V3 serving on TPU pod slices — to within rounding error, indicating that design docs, rather than code, can be the durable artifact for ML system co-design tools.
--- *Auto-collected on 2026-09-09.*