Version 1.1.2
Download NEURAX
NEURAX is available on Linux now. We are working on the macOS and Windows builds; each will be published here when ready. No account, and nothing written outside your home directory.
Currently, NEURAX’s full workflow through training works correctly. The team is working to stabilize it and make training more reliable and robust.
Every package
Linux
Debian, Ubuntu, Kali, Arch, Fedora, and effectively any distribution.
macOS
Intel and Apple silicon, one universal build.
We are working on this build now. We will publish it here as soon as it is ready.
Windows
Windows 10 and 11. The installer adds the WebView2 runtime on Windows 10 if it is missing.
We are working on this build now. We will publish it here as soon as it is ready.
Every file above is attached to this release. SHA256SUMS.txt (in the release itself) lists a checksum for each one.
The machine NEURAX needs
These four profiles are guidance, not hard limits. NEURAX can design and analyze on a wider range of machines; whether a training run fits and finishes depends on the supported runtime, architecture, dataset and hardware. The profiles describe common setups. On a weaker machine the same work may be slow or may not fit; review the Studio's estimate before approval and decide whether to proceed.
Minimum
- Processor
- 4 cores, 64-bit (x86-64 with AVX2, or Apple Silicon)
- Memory
- 8 GB
- GPU
- None
- Disk
- 20 GB free
Trains comfortably
Small models, up to about 10 million parameters, on a few thousand samples.
Warm start (from v0.24)
Compatible export from an earlier NEURAX run; confirm the model, data and memory fit before restarting.
Standard CPU
- Processor
- 8 cores, or Apple M1 and newer
- Memory
- 16 GB
- GPU
- None
- Disk
- 50 GB on an SSD
Trains comfortably
Models up to about 25 million parameters, on up to about 50,000 samples.
Warm start (from v0.24)
Compatible export from an earlier NEURAX run; confirm the architecture, data and available RAM.
Standard GPU
- Processor
- 8 cores
- Memory
- 32 GB
- GPU
- NVIDIA RTX 3060 12 GB, RTX 4070 or newer
- Disk
- 100 GB on an SSD
Trains comfortably
Models up to about 500 million parameters.
Warm start (from v0.24)
Compatible export from an earlier NEURAX run; confirm the architecture, batch size and available VRAM.
Advanced GPU
- Processor
- 16 cores
- Memory
- 64 GB
- GPU
- NVIDIA with 24 GB or more: RTX 3090, 4090, 5090, L4, A100, H100
- Disk
- 200 GB on an SSD
Trains comfortably
Models up to about 1 billion parameters.
Warm start (from v0.24)
Compatible export from an earlier NEURAX run; confirm the model, data, batch size and available VRAM.
- GPUs: NVIDIA only for training today, Turing (RTX 20xx, GTX 16xx, T4) or newer, with a driver supporting CUDA 12.8. AMD and Intel GPUs, and Apple's GPU, are detected but not yet used: a Mac trains on its processor.
- The one run NEURAX does not start is one whose memory cannot fit the machine at all, since it would crash; it tells you what would fit instead.
- From NEURAX 0.24, a supported run can continue from an earlier NEURAX run's export. This does not import or fine-tune Hugging Face or ModelScope checkpoint weights; check compatibility with Studio analysis and a small pilot.
Before you install
Do I need to know machine learning?
No. You need to know your problem and your data. Choosing an architecture, sizing it, writing the PyTorch and fitting it to your hardware is the work NEURAX does.
Does it run my model to measure it?
No. It compiles the architecture to the operations it's made of and prices them, which is why it answers in milliseconds, and why it can cost a model far larger than your machine.
Does my data leave my machine?
The Studio reads supported datasets locally for profiling and training. If you use the optional Agent, your chosen provider receives its prompt and project design/data profile; NEURAX says it does not send raw dataset contents or the local path. Importing a public Hugging Face or ModelScope model also makes an online request for its configuration.
Do I need a GPU?
Not to design, cost or compare models: that runs on a CPU. You need one to train quickly, and NEURAX tells you what a run costs on the machine you have, including a laptop.
Can it really train, or only plan?
It trains. The run is written into your project folder as real files, runs on your machine, reports every step live, and checkpoints so it can be paused and resumed.
What do I own at the end?
Ordinary PyTorch files and trained weights, in your own folder. Nothing is locked to NEURAX.
Can we use our own internal LLM?
Yes. The assistant talks to whatever endpoint you configure, including a model hosted inside your own network.
What does it cost?
Community is free. Developer is $4.99 a month and Pro is $20 a month; Enterprise is priced with you. The assistant runs on your own provider key, so its model usage is whatever your provider charges.