Complete neural network benchmarks with Reactant, JAX, and PyTorch#1530
Merged
ChrisRackauckas merged 3 commits intoSciML:masterfrom Mar 27, 2026
Merged
Complete neural network benchmarks with Reactant, JAX, and PyTorch#1530ChrisRackauckas merged 3 commits intoSciML:masterfrom
ChrisRackauckas merged 3 commits intoSciML:masterfrom
Conversation
- Rebase onto master and update to latest package versions (Lux 1.31, Reactant 0.2.240, Flux 0.16) - Add Reactant.jl inference (via @compile) and training (via TrainState API) - Add JAX and PyTorch benchmarks via PythonCall with timing done entirely in Python - Complete all benchmark sections: MLP relu, MLP gelu, MLP+BatchNorm, LeNet 5, Small ResNet - Add both inference and training benchmarks for all model types - Configure GPU runner via benchmark_config.toml - Add CondaPkg.toml for Python dependencies (jax, torch, optax) - Add .CondaPkg and __pycache__ to .gitignore Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com> Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
20 tasks
ChrisRackauckas-Claude
pushed a commit
to ChrisRackauckas-Claude/SciMLBenchmarksOutput
that referenced
this pull request
Mar 27, 2026
Prepare output directory structure for the new NeuralNetworks benchmark (SciML/SciMLBenchmarks.jl#1530) which compares Lux, Flux, SimpleChains, Reactant, JAX, and PyTorch on common neural network workloads. Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com> Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2 tasks
Add "Neural Network Framework Benchmarks" section to the ordered docs navigation, placed after the PINN benchmarks. Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com> Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Completes the work started in #1026 by @avik-pal. This PR:
@compile) and training (viaTrainState+AutoEnzymeAPI)benchmark_config.toml(["self-hosted", "gpu", "exclusive"])CondaPkg.tomlfor Python dependencies (jax, jaxlib, optax, torch)Frameworks compared
Architecture
nn_benchmark_utils.py(imported via PythonCall)time.perf_counter()inside Python loops to avoid measuring Julia↔Python call overheadblock_until_ready()for accurate async timingTested locally on CPU
All 6 frameworks verified working for inference and training on CPU (no GPU available for local testing).
Supersedes #1026
Test plan
🤖 Generated with Claude Code