🛣️ AioWay
Aioway is an optimizing compiler for deep learning algorithms. It treats the machine learning models / algorithms as instructions and build the pipeline that way.
🏢 Architecture diagram
🍰 Features
Most of these features are done but not polished yet! But will be in couple of weeks.
- ⚡ Fast. Compared to neural architecture search, the optimization can be rule based (fast).
- 🕵️ Detects the tasks at hand, resource available, and select the best algorithms and models.
- 🎁 The models built from aioway would be white box (explainable), due to our architecture.
- ⚙️ Allows upgrading parts of the models. You scale up to different model size, and to different machines.
- 🐍 Both relational algebra (SQL like) and python library interface.
- 🔥 Extensible with custom pytorch.
⛰️ Compared with current landscape
- Neural architecture search: Too slow (because most need backtracking).
- Current autoML framework: non flexible enough, usually stuck with UI or fixed set of models (black box still).
- Pretrained models / LLM: Usually expensive, non explainable, less flexible.
- Traditional methods: They can't handle new data (multimodal).
🤔 Why aioway yada yada
In the recent years, machine learning's entry barrier higher, rather than lower. People with expert training are expensive, as they need years of experience to be good.
However, current AutoML solutions are subpar. Each one of them have clear limitations: slow, inflexible, unreliable, or unable to handle modern data.
Drawing inspriation from opitmizing compilers (especially SQLs), aioway aims to solve that.
🌟 Give us a star!
That's all for now!
If you have read this far, please consider giving me a star (⭐) or a fork (🍴).
This will keep me motivated!
Or if you have too much cash at hand:
🗺️ Roadmap
We are most likely launching v0.1.0 before July 2026, but before that,
see the pre-release tracking project for more details.
🤝 Contributing
Contributing is of course welcome. Please see the contributing guide and follow the code of conduct.
👨👨👦👦 Contributors
🐨 Relation to koila
Aioway builds on top of the original koila (moved to a branch). The torch team built FakeTensor which overlaps a lot with koila's functionality, so it's no longer maintained. See the rationale in the koila branch.
Conceptually, aioway works in a similar way, but instead of Tensor ops, aioway focuses on a higher level, on algorithm building.
1 maintainer
Metadata
1 maintainer
Release files for aioway 0.3.3
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| aioway-0.3.3.tar.gz | 251.4 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| aioway-0.3.3-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 384.2 kB
Release files / aioway-0.3.3.tar.gz
| Download URL | aioway-0.3.3.tar.gz |
|---|---|
| Size | 251.4 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
ef63f548d1c2865b4c3a2a224d6d0a230c880fcd56f1784be3caba5ec5ae2c77
|
|
BLAKE2b-256 checksum How to use checksums |
8a4b4df33085b2b6e5b2e14687b8a5b1171e72e400d0ba8e08edf7fd2f8b5c32
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
pdm/2.29.2 CPython/3.14.8 Linux/6.17.0-1022-azure
|
Release files / aioway-0.3.3-py3-none-any.whl
| Download URL | aioway-0.3.3-py3-none-any.whl |
|---|---|
| Size | 132.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
eff281bb48d2ec2604b7bd858afca9e70b3ff1f4c517985397c8d24678fb1bdf
|
|
BLAKE2b-256 checksum How to use checksums |
1f5e2ccebf38bd8af7419d745e80298494dadb77795da50508c6f54fac941446
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
pdm/2.29.2 CPython/3.14.8 Linux/6.17.0-1022-azure
|
