mirror of https://github.com/harvard-edge/cs249r_book.git synced 2026-05-06 09:38:33 -05:00

Files

Vijay Janapa Reddi 4309c3f7ee mlsysim(CHANGELOG): restructure 0.1.1 entry to match release-notes template

Rewrite the 0.1.1 changelog entry to use the 9-category structure that
release notes are organized around (Solvers, Hardware, Workloads, CLI,
Python API, Documentation, Packaging, Bug Fixes, Internal — only
sections with content shown). Adds an "—Optional Theme" extension
to the heading so future releases can carry a short descriptor.

Adds a format-note comment block at the top of CHANGELOG.md documenting
the structure so future authors write entries in the shape the
mlsysim-pypi-publish workflow expects to extract.

Leaves the v0.1.0 entry untouched (its release notes were hand-written
via RELEASE_NOTES_0.1.0.md, so extraction doesn't apply).

2026-04-24 16:01:25 -04:00

5.7 KiB

Raw Permalink Blame History

Changelog

v0.1.1 (2026-04-24) — Paper Title Correction

Metadata-only patch release. No code or API changes; safe drop-in replacement for 0.1.0. Corrects the paper title cited in three places to match the actual title of the companion paper.

📚 Documentation

Paper title corrected across CITATION.cff, the BibTeX snippet in README.md, and the reference docstring in mlsysim/core/walls.py. Was: "A Composable Analytical Framework for Machine Learning Systems." Now: "MLSys·im: First-Principles Infrastructure Modeling for Machine Learning Systems."

🏗️ Packaging & Dependencies

Version bumped to 0.1.1 across pyproject.toml, mlsysim/__init__.py, and CITATION.cff; date-released updated to 2026-04-24.

Contributors

@profvjreddi

v0.1.0 (2026-04-01)

Initial release of MLSysim — the first-principles analytical modeling engine for ML systems.

Core Framework

22-wall taxonomy organizing every constraint that bounds ML system performance
20+ analytical solvers (Models, Solvers, Optimizers) covering all 22 walls
Pint unit system with dimensional analysis throughout
TraceableConstant pattern — key defaults carry citations for every assumption
Pipeline composer for chaining solvers with explain() and run()
3-tier evaluation scorecard: Feasibility → Performance → Macro/Economics
Design Space Exploration (DSE) engine with constraint evaluation

Hardware Registry

15+ accelerators: V100, A100, H100, H200, B200, GB200 NVL72, MI300X, TPUv5p, T4, Cerebras CS-3, Jetson Orin NX, ESP32-S3, nRF52840, Himax WE-I Plus, DGX Spark, MacBook M3 Max, iPhone 15 Pro, Pixel 8
Full precision support: FP32, TF32, BF16, FP16, FP8, INT8, INT4
Multi-level memory hierarchy: HBM + SRAM + Flash (TinyML)
All specs verified against manufacturer datasheets

Model Registry

GPT-2/3/4, LLaMA-2/3 (7B/8B/70B), BERT Base/Large, ResNet-50, MobileNetV2, AlexNet, Mamba, Stable Diffusion v1.5, DS-CNN, WakeVision
HuggingFace model importer for custom workloads

Analytical Models

SingleNodeModel: Roofline analysis with SRAM/flash-aware bandwidth selection
DistributedModel: 4D parallelism (DP/TP/PP/EP) with correct activation-based TP communication, gradient accumulation, straggler effects
ServingModel: Prefill/decode with attention O(S²), batch amortization, speculative decoding, disaggregated serving
ContinuousBatchingModel: PagedAttention with KV cache compression
WeightStreamingModel: Cerebras-style with prefill/decode phases
TailLatencyModel: Erlang C (M/M/c) with log-space computation for large clusters
ReliabilityModel: Compound MTBF with correlated failures, goodput ratio
CheckpointModel: Distributed writing with filesystem bandwidth limits
SustainabilityModel: Energy-proportional power, embodied carbon, PUE/WUE/carbon intensity
EconomicsModel: Amortized CapEx, infrastructure multiplier, maintenance
CompressionModel: Quantization (FP8/INT8/INT4) + pruning (unstructured/structured/N:M) with inference speedup
ScalingModel: Chinchilla compute-optimal scaling
TopologyModel: Ring, torus, fat-tree, dragonfly bisection analysis
SensitivitySolver: Numerical partial derivatives for binding constraint identification
SynthesisSolver: Inverse Roofline for hardware spec derivation
ParallelismOptimizer, BatchingOptimizer, PlacementOptimizer: Design-space search

CLI

mlsysim eval — evaluate workload on hardware
mlsysim zoo — explore hardware/model registries
mlsysim schema — export solver schemas
mlsysim optimize — design-space search
mlsysim audit — system audit

Testing

367 tests, 100% pass rate
Direct formula unit tests with known-answer validation
Solver suite covering all major models
Physics bounds validation across all hardware
Wall taxonomy completeness tests
Pipeline composition tests

Packaging & Tooling

Standard nested package layout (mlsysim/mlsysim/...) so pip install -e . works out of the box without the prefix-add sources rewrite that broke the editables backend
Wheel and sdist contain only the package and project metadata (no tests/, docs/, examples/, paper/, vscode-ext/)
Project-wide ruff configuration: [tool.ruff] block in pyproject.toml with sensible per-file ignores for __init__.py re-export patterns, core/constants.py unit-registry star imports, and tests/examples idioms; ruff check . reports zero issues
Real bug fixes uncovered by lint cleanup: removed unused BaseModel import in core/solver.py, fixed Fleet shadowing bug in sim/simulations.py, narrowed three bare except: clauses to specific exception types, and added missing speculative-decoding ITL assertion in tests/test_sota.py

5.7 KiB Raw Permalink Blame History