Design Docs Are All You Need: SMART, an AI-Native ML Performance Modeling Library

Forum topic · 小凯 · 2026-09-09

Summary

This arXiv paper (2609.05364) introduces SMART, a symbolic performance-modeling library for machine learning systems built around an AI-native development model. The main branch contains almost no code: the repository is a DAG of self-contained natural-language design docs, and coding sub-agents regenerate the entire implementation from the docs on each version update, with every human change being a natural-language doc edit. Two design elements make regeneration reliable: design docs structured around step-by-step worked examples that serve as in-context demonstrations for the generating agents, and a minimal, recursively defined operational IR with symbolic (SymPy) cost expressions, a fast analytical summary mode for large-scale sweeps, and a slow module-scheduling mode for fine-grained scheduling studies. The regenerated implementations reproduce hand-audited reference models, including DeepSeek-V3 serving on TPU pod slices, to rounding precision, suggesting that design documents rather than code can serve as the durable artifact for ML system co-design.

Overview

Field: Machine Learning 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 technical debt of incrementally patching it.

The 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 — the system is self-documenting.

Key design elements

1. Doc style built around step-by-step worked examples, which act as in-context demonstrations for the generating agents. 2. A minimal, recursively defined operational IR featuring:

Symbolic (SymPy) cost expressions,
A fast analytical summary mode for large-scale sweeps,
A slow module-scheduling mode for fine-grained scheduling studies.

Results

The regenerated implementations reproduce hand-audited reference models — including DeepSeek-V3 serving on TPU pod slices — to rounding precision. This suggests that design docs, rather than code, can be the durable artifact for ML system co-design.

Original Abstract

> Machine-learning performance modeling is a uniquely hostile terrain for long-lived software: the 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...

--- *Auto-collected on 2026-09-09*

Design Docs Are All You Need: SMART, an AI-Native ML Performance Modeling Library

Summary

Overview

Summary

Key design elements

Results

Original Abstract

Tags