diffrax
FrameworkBenchmarks
Our great sponsors
diffrax | FrameworkBenchmarks | |
---|---|---|
21 | 366 | |
1,230 | 7,384 | |
- | 1.2% | |
8.3 | 9.8 | |
6 days ago | 4 days ago | |
Python | Java | |
Apache License 2.0 | GNU General Public License v3.0 or later |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
diffrax
- Ask HN: What side projects landed you a job?
-
[P] Optimistix, nonlinear optimisation in JAX+Equinox!
Optimistix has high-level APIs for minimisation, least-squares, root-finding, and fixed-point iteration and was written to take care of these kinds of subroutines in Diffrax.
-
Show HN: Optimistix: Nonlinear Optimisation in Jax+Equinox
Diffrax (https://github.com/patrick-kidger/diffrax).
Here is the GitHub: https://github.com/patrick-kidger/optimistix
The elevator pitch is Optimistix is really fast, especially to compile. It
-
Scientific computing in JAX
Sure. So I've got some PyTorch benchmarks here. The main take-away so far has been that for a neural ODE, the backward pass takes about 50% longer in PyTorch, and the forward (inference) pass takes an incredible 100x longer.
-
[D] JAX vs PyTorch in 2023
FWIW this worked for me. :D My full-time job is now writing JAX libraries at Google. Equinox for neural networks, Diffrax for differential equation solvers, etc.
-
Returning to snake's nest after a long journey, any major advances in python for science ?
It's relatively early days yet, but JAX is in the process of developing its nascent scientific computing / scientific machine learning ecosystem. Mostly because of its strong autodifferentiation capabilities, excellent JIT compiler etc. (E.g. to show off one of my own projects, Diffrax is the library of diffeq solvers for JAX.)
-
What's the best thing/library you learned this year ?
Diffrax - solving ODEs with Jax and computing it's derivatives automatically functools - love partial and lru_cache fastprogress - simpler progress bar than tqdm
-
PyTorch 2.0
At least prior to this announcement: JAX was much faster than PyTorch for differentiable physics. (Better JIT compiler; reduced Python-level overhead.)
E.g for numerical ODE simulation, I've found that Diffrax (https://github.com/patrick-kidger/diffrax) is ~100 times faster than torchdiffeq on the forward pass. The backward pass is much closer, and for this Diffrax is about 1.5 times faster.
It remains to be seen how PyTorch 2.0 will compare, or course!
Right now my job is actually building out the scientific computing ecosystem in JAX, so feel free to ping me with any other questions.
-
Python 3.11 is much faster than 3.8
https://github.com/patrick-kidger/diffrax
Which are neural network and differential equation libraries for JAX.
[Obligatory I-am-googler-my-opinions-do-not-represent- your-employer...]
-
Ask HN: What's your favorite programmer niche?
Autodifferentiable programming!
Neural networks are the famous example of this, of course -- but this can be extended to all of scientific computing. ODE/SDE solvers, root-finding algorithms, LQP, molecular dynamics, ...
These days I'm doing all my work in JAX. (E.g. see Equinox or Diffrax: https://github.com/patrick-kidger/equinox, https://github.com/patrick-kidger/diffrax). A lot of modern work is now based around hybridising such techniques with neural networks.
I'd really encourage anyone interested to learn how JAX works under-the-hood as well. (Look up "autodidax") Lots of clever/novel ideas in its design.
FrameworkBenchmarks
-
Why choose async/await over threads?
Neat. Thanks for sharing!
Interestingly, may-minihttp is faring very well in the TechEmpower benchmark [1], for whatever those benchmarks are worth. The code is also surprisingly straightforward [2].
[1] https://www.techempower.com/benchmarks/
[2] https://github.com/TechEmpower/FrameworkBenchmarks/blob/mast...
-
Ntex: Powerful, pragmatic, fast framework for composable networking services
ntex was formed after a schism in actix-web and Rust safety/unsafety, with ntex allowing more unsafe code for better performance.
ntex is at the top of the TechEmpower benchmarks, although those benchmarks are not apples-to-apples since each uses its own tricks: https://www.techempower.com/benchmarks/#hw=ph&test=fortune&s...
-
A decent VS Code and Ruby on Rails setup
Ruby is slow. Very slow. How much you may ask? https://www.techempower.com/benchmarks/#hw=ph&test=fortune&s... fastest Ruby entry is at 272th place. Sure, top entries tend to have questionable benchmark-golfing implementations, but it gives you a good primer on the overhead imposed by Ruby.
It is also not early 00s anymore, when you pick an interpreted language, you are not getting "better productivity and tooling". In fact, most interpreted languages lag behind other major languages significantly in the form of JS/TS, Python and Ruby suffering from different woes when it comes to package management and publishing. I would say only TS/JS manages to stand apart with being tolerable, and Python sometimes too by a virtue of its popularity and the amount of information out there whenever you need to troubleshoot.
If you liked Go but felt it being a too verbose to your liking, give .NET a try. I am advocating for it here on HN mostly for fun but it is, in fact, highly underappreciated, considered unsexy and boring while it's anything but after a complete change of trajectory in the last 3-5 years. It is actually the* stack people secretly want but simply don't know about because it is bundled together with Java in the public perception.
*productive CLI tooling, high performance, works well in a really wide range of workloads from low to high level, by far the best ORM across all languages and back-end framework that is easier to work with than Node.JS while consuming 0.1x resources
-
The Erlang Ecosystem [video]
Although that seems to have improved in recent years.
https://www.techempower.com/benchmarks/#hw=ph&test=json§...
-
Ruby 3.3
RoR and whatever C++ based web backend there is count as a valid comparison in my book. But comparing the languages itself is maybe a bit off.
On a side note, you can actually compare their performance here if you’re really curious. But take it with a grain of salt since these are synthetic benchmarks.
https://www.techempower.com/benchmarks
-
API: Go, .NET, Rust
Most benchmarks you'll find essentially have someone's thumb on the scale (intentionally or unintentionally). Most people won't know the different languages well enough to create comparable implementations and if you let different people create the implementations, cheating happens. The TechEmpower benchmarks aren't bad, but many implementations put their thumb on the scale (https://www.techempower.com/benchmarks). For example, a lot of the Go implementations avoid the GC by pre-allocating/reusing structs or allocate arrays knowing how big they need to be in advance (despite that being against the rules). At some point, it becomes "how many features have you turned off." Some Go http routers (like fasthttp and those built off it like Atreugo and Fiber) aren't actually correct and a lot of people in the Go community discourage their use, but they certainly top the benchmarks. Gin and Echo are usually the ones that are well-respected in the Go community.
-
Rage: Fast web framework compatible with Rails
There is certainly a lot of speculation in Techempower benchmarks and top entries can utilize questionable techniques like simply writing a byte array literal to output stream instead of constructing a response, or (in the past) DB query coalescing to work around inherent limitations of the DB in case of Fortunes or DB quries.
And yet, the fastest Ruby entry is at 274th place while Rails is at 427th.
https://www.techempower.com/benchmarks/#hw=ph&test=fortune&s...
-
Node.js – v20.8.1
oh what machine? with how many workers? doing what?
search for "node" on this page: https://www.techempower.com/benchmarks/#section=data-r21
-
Strong typing, a hill I'm willing to die on
JustJS would like a word https://www.techempower.com/benchmarks/#section=data-r20&tes...
-
Rust vs Go: A Hands-On Comparison
In terms of RPS, this web service is more-or-less the fortunes benchmark in the techempower benchmarks, once the data hits the cache: https://www.techempower.com/benchmarks/#section=data-r21
Or, at least, they would be after applying optimizations to them.
In short, both of these would serve more rps than you will likely ever need on even the lowest end virtual machines. The underlying API provider will probably cut you off from querying them before you run out of RPS.
What are some alternatives?
deepxde - A library for scientific machine learning and physics-informed learning
zio-http - A next-generation Scala framework for building scalable, correct, and efficient HTTP clients and servers
tiny-cuda-nn - Lightning fast C++/CUDA neural network framework
drogon - Drogon: A C++14/17 based HTTP web application framework running on Linux/macOS/Unix/Windows [Moved to: https://github.com/drogonframework/drogon]
flax - Flax is a neural network library for JAX that is designed for flexibility.
django-ninja - 💨 Fast, Async-ready, Openapi, type hints based framework for building APIs
juliaup - Julia installer and version multiplexer
LiteNetLib - Lite reliable UDP library for Mono and .NET
equinox - Elegant easy-to-use neural networks + scientific computing in JAX. https://docs.kidger.site/equinox/
C++ REST SDK - The C++ REST SDK is a Microsoft project for cloud-based client-server communication in native code using a modern asynchronous C++ API design. This project aims to help C++ developers connect to and interact with services.
dm-haiku - JAX-based neural network library
SQLBoiler - Generate a Go ORM tailored to your database schema.