NeurEco 5.0.2052 Python library =============================== Contents -------- neureco-5.0.2052+torch2.9.cu129-py3-none-manylinux_2_35_x86_64.whl Linux, PyTorch 2.9, NVIDIA GPU (CUDA 12), 119.4 MB neureco-5.0.2052+torch2.9.cpu-py3-none-manylinux_2_35_x86_64.whl Linux, PyTorch 2.9, CPU only, 119.1 MB neureco-5.0.2052+torch2.8.cu129-py3-none-manylinux_2_35_x86_64.whl Linux, PyTorch 2.8, NVIDIA GPU (CUDA 12), 119.3 MB neureco-5.0.2052+torch2.7.cu126-py3-none-manylinux_2_35_x86_64.whl Linux, PyTorch 2.7, NVIDIA GPU (CUDA 12), 133.2 MB neureco-5.0.2052+torch2.9.cu128-py3-none-win_amd64.whl Windows, PyTorch 2.9, NVIDIA GPU (CUDA 12.8), 96.7 MB neureco-5.0.2052+torch2.9.cpu-py3-none-win_amd64.whl Windows, PyTorch 2.9, CPU only, 96.4 MB neureco-5.0.2052+torch2.8.cu128-py3-none-win_amd64.whl Windows, PyTorch 2.8, NVIDIA GPU (CUDA 12.8), 96.6 MB neureco-5.0.2052+torch2.7.cu126-py3-none-win_amd64.whl Windows, PyTorch 2.7, NVIDIA GPU (CUDA 12.6), 96.9 MB cpu/ package index for pip: CPU only, PyTorch 2.9 torch2.9/ package index for pip: NVIDIA GPU, PyTorch 2.9 torch2.8/ package index for pip: NVIDIA GPU, PyTorch 2.8 torch2.7/ package index for pip: NVIDIA GPU, PyTorch 2.7 SHA256SUMS SHA-256 checksums of these files README.txt this file Choosing an edition ------------------- Like PyTorch, the NeurEco Python library comes in several editions: a CPU-only edition and NVIDIA GPU editions, one per PyTorch version. Each wheel is built for one system (Linux or Windows, x86-64) and one PyTorch minor version, and requires it. Choose the edition for your hardware (CPU only or NVIDIA GPU) and for the PyTorch version you use: pip takes the wheel for your system from the package index of that edition, and installs that PyTorch version if needed, together with NumPy (1.24 or newer) and Matplotlib. Install one edition per Python environment: every edition is named neureco. pip keeps the neureco and torch packages already installed when their versions fit, even if they belong to another edition or if PyTorch is a CPU-only build: to change edition, first run pip uninstall neureco torch, or use a new virtual environment. If the PyTorch installed in your environment is not the version the wheel was built for, NeurEco shows a warning when it is imported. Requirements ------------ - Python 3.10 or newer, in a version that your PyTorch version supports: 3.10 to 3.14 for PyTorch 2.9, 3.10 to 3.13 for PyTorch 2.7 and 2.8. - Linux: glibc 2.35 or newer and the C++ runtime of GCC 12 or newer, as in Ubuntu 22.04 or later, Debian 12 or later and Red Hat Enterprise Linux 10 or later. pip does not install the library on Red Hat Enterprise Linux 8 or 9 and compatible systems. - NVIDIA GPU editions: compute capability 7.5 or higher, with NVIDIA driver 525.60.13 (Linux) or 527.41 (Windows) or newer (575.51.03 or 576.02 or newer recommended). GeForce RTX 50 series and other GPUs of compute capability 10.0 or higher need the PyTorch 2.8 or 2.9 edition. Python environment ------------------ On Linux, install NeurEco in a virtual environment: Debian 12, Ubuntu 23.04 and later refuse pip installs into the system Python. For example: python3 -m venv ~/neureco-env . ~/neureco-env/bin/activate On Debian and Ubuntu, python3 -m venv needs the package python3-venv. Run the commands below in that environment. Upgrading from NeurEco 4.x -------------------------- The INSTALL.py script of NeurEco 4.x copied the Python library into the standard-library folder of your Python installation, where Python finds it before the packages that pip installs, even in a virtual environment. Before you install, check which copy Python finds: python -c "import importlib.util as u; s = u.find_spec('NeurEco'); print(s.origin if s else 'not found')" If it prints not found, no copy is installed. If it prints a path that is not in a site-packages folder (dist-packages for the system Python of Debian and Ubuntu), delete that NeurEco folder (this can need administrator rights) and run the command again. Once NeurEco 5.0 is installed, the command must print a path in site-packages or dist-packages. Installation ------------ Install your edition with one pip command, the same on Linux and Windows: pip takes NeurEco from the Adagos package index of the edition, PyTorch from the PyTorch package index, and NumPy and Matplotlib from PyPI. CPU only, PyTorch 2.9: pip install "neureco~=5.0" --extra-index-url https://download.pytorch.org/whl/cpu --extra-index-url https://adagos.com/downloads/neureco5/whl/cpu/ NVIDIA GPU, PyTorch 2.9: pip install "neureco~=5.0" --extra-index-url https://download.pytorch.org/whl/cu128 --extra-index-url https://adagos.com/downloads/neureco5/whl/torch2.9/ NVIDIA GPU, PyTorch 2.8: pip install "neureco~=5.0" --extra-index-url https://download.pytorch.org/whl/cu128 --extra-index-url https://adagos.com/downloads/neureco5/whl/torch2.8/ NVIDIA GPU, PyTorch 2.7: pip install "neureco~=5.0" --extra-index-url https://download.pytorch.org/whl/cu126 --extra-index-url https://adagos.com/downloads/neureco5/whl/torch2.7/ The PyTorch package index provides the PyTorch build of the edition: on Windows, PyTorch on PyPI is CPU-only, and on Linux, it includes CUDA, which makes a large download that the CPU-only edition does not need. Keep "neureco~=5.0" as written: PyPI has a placeholder package named neureco, which pip would install whenever it cannot use the Adagos package: when it cannot read the Adagos package index, or when your Python version or system is not supported. With ~=5.0, pip stops with an error instead. Behind a proxy or a firewall, pip needs adagos.com, pypi.org, files.pythonhosted.org, download.pytorch.org, download-r2.pytorch.org and, for the NVIDIA GPU editions on Linux, pypi.nvidia.com. If pip cannot reach the PyTorch package index, it takes PyTorch from PyPI without an error, which on Windows gives the CPU-only build. Use one Adagos package index at a time, on the command line and in your pip configuration: with several, pip can install an edition other than the one you chose, for example a GPU edition with a CPU-only PyTorch. To update NeurEco to a newer build, run the command of your edition with --upgrade. Do not use --upgrade to change edition: it can replace neureco with another edition and keep the installed PyTorch build. The package index of each edition holds only the latest build of NeurEco 5.0: once a newer build is published, pip can no longer install an earlier one from it. A requirement that names a build, such as the line neureco==5.0.2052+torch... that pip freeze writes, then fails with "No matching distribution found". In requirement files, write "neureco~=5.0" instead, or keep the wheel file of the build you need and install it as described below. To run exported ONNX models, install onnxruntime or onnxruntime-gpu yourself. Installing a downloaded wheel ----------------------------- To install a wheel file of this folder that you downloaded, run the commands for that wheel in the folder that holds it. On Linux, PyTorch on PyPI already includes CUDA. On Windows, PyTorch on PyPI is CPU-only, so the GPU editions first install the CUDA build of PyTorch. Linux, PyTorch 2.9, NVIDIA GPU (CUDA 12): pip install ./neureco-5.0.2052+torch2.9.cu129-py3-none-manylinux_2_35_x86_64.whl Linux, PyTorch 2.9, CPU only: pip install "torch==2.9.*" --index-url https://download.pytorch.org/whl/cpu pip install ./neureco-5.0.2052+torch2.9.cpu-py3-none-manylinux_2_35_x86_64.whl Linux, PyTorch 2.8, NVIDIA GPU (CUDA 12): pip install ./neureco-5.0.2052+torch2.8.cu129-py3-none-manylinux_2_35_x86_64.whl Linux, PyTorch 2.7, NVIDIA GPU (CUDA 12): pip install ./neureco-5.0.2052+torch2.7.cu126-py3-none-manylinux_2_35_x86_64.whl Windows, PyTorch 2.9, NVIDIA GPU (CUDA 12.8): pip install "torch==2.9.*" --index-url https://download.pytorch.org/whl/cu128 pip install .\neureco-5.0.2052+torch2.9.cu128-py3-none-win_amd64.whl Windows, PyTorch 2.9, CPU only: pip install .\neureco-5.0.2052+torch2.9.cpu-py3-none-win_amd64.whl Windows, PyTorch 2.8, NVIDIA GPU (CUDA 12.8): pip install "torch==2.8.*" --index-url https://download.pytorch.org/whl/cu128 pip install .\neureco-5.0.2052+torch2.8.cu128-py3-none-win_amd64.whl Windows, PyTorch 2.7, NVIDIA GPU (CUDA 12.6): pip install "torch==2.7.*" --index-url https://download.pytorch.org/whl/cu126 pip install .\neureco-5.0.2052+torch2.7.cu126-py3-none-win_amd64.whl License ------- Running NeurEco requires a license for NeurEco 5.0: license files issued for NeurEco 4.x do not work with it. To request one, run python -m NeurEco.hardware_info It asks for your contact details and writes the file hardware_info.txt in the current folder. Send hardware_info.txt to support@adagos.com to obtain your license. On most virtual machines, which include most cloud servers, NeurEco does not create hardware_info.txt: to use NeurEco on a virtual machine, contact support@adagos.com. Then set the environment variable ADAGOS_LICENSE_FILE to the full path of the license file. On Windows, you can instead copy the license file, with its .lic extension, to your user folder (usually C:\Users\). On Linux, without the variable, NeurEco finds the license file only in the current folder. Checking the files ------------------ On Linux, run this in the folder of this file: sha256sum -c SHA256SUMS --ignore-missing On Windows, run this in PowerShell, in the folder of this file. Every line must end with OK: Get-Content SHA256SUMS | ForEach-Object { $h, $f = $_ -split '\s+', 2; if (Test-Path -LiteralPath $f) { "$f " + $(if ((Get-FileHash -LiteralPath $f).Hash -eq $h) { 'OK' } else { 'MISMATCH' }) } } NeurEco 5.0.2052, build 338 (78e46ea2405f), built on 2026-11-02.