Installing SLEAP in a computer without internet

[Saheli-R]

Hi,

I have successfully installed SLEAP in a computer, but now want to transfer the files from this computer to another desktop without internet connection, via external hard drives. I want to use this second computer for training the model. I already have anaconda working in this second computer, so I am thinking it should be possible to do such an offline installation. But could you please guide me with the exact steps I would need to follow?

Thanks!

[talmo]

Hi @Saheli-R,

This is possible but a bit tricky -- do you mind telling us a bit more about the systems?

1. What is the Operating System and version on the current computer vs the one without internet?
2. Do either of them have GPUs?

Cheers,

Talmo

[Saheli-R]

Hi Talmo,

1. Both are Windows computers. The one where I already have SLEAP installed is running Windows 10 Pro; the one without the internet is running Windows 11.

2. Both have GPUs - the current computer has NVIDIA GeForce GTX 1050 Ti; the other one has Zotac GeForce RTX 4080 16GB trinity.

In my current computer SLEAP is running slowly and I want to process several videos, each 90 minutes long. So I am hoping the other computer will be much faster.

Let me know if you need any other information, and thanks for your help!

[talmo]

Hi @Saheli-R,

Thanks for the post! Great question — offline installation is definitely doable.

We've put together a pre-built offline installer for SLEAP on Windows with NVIDIA GPU support. There are two options depending on your setup:

---

Option A: Use our pre-built pack (easiest)

We've packaged everything you need into a single zip file — Python, SLEAP, PyTorch with CUDA, and all dependencies.

Download: sleap-offline-pack-win-cu128.zip (~3.4 GB)

On the offline computer

1. Extract the zip

Copy the zip to the offline computer via USB drive. Extract it to any folder you like — your Desktop, Documents, or anywhere else. Right-click the zip file and select "Extract All...", or open Command Prompt and run:

tar -xf D:\sleap-offline-pack-win-cu128.zip -C %USERPROFILE%\Desktop

(Replace D:\ with the drive letter of your USB, and %USERPROFILE%\Desktop with wherever you'd like to put it.)

You should now have a pack\ folder with files like unpack.cmd, wheels\, python\, etc.

2. Run the installer

Open Command Prompt, navigate to wherever you extracted the zip, and run:

cd %USERPROFILE%\Desktop
pack\unpack.cmd

This takes a few minutes. It will:

• Extract a bundled copy of Python 3.13
• Create a virtual environment
• Install all 185+ packages from local files (no internet needed)

3. Activate and verify

pack\.venv\Scripts\activate.bat
sleap doctor

This prints a diagnostic report showing your SLEAP version, GPU status, and installed packages. Look for SLEAP: 1.6.1 and CUDA: available in the output.

4. Launch SLEAP

pack\.venv\Scripts\activate.bat
sleap label

You'll need to run the activate.bat line each time you open a new Command Prompt window before using SLEAP.

PowerShell instructions (click to expand)

If you prefer PowerShell over Command Prompt:

# Extract (change the paths to wherever you like)
tar -xf D:\sleap-offline-pack-win-cu128.zip -C $HOME\Desktop
cd $HOME\Desktop

# Unpack
& pack\unpack.ps1

# Activate
& pack\.venv\Scripts\Activate.ps1

# Verify
sleap doctor

# Launch
sleap label

---

Option B: Build the pack yourself (if you need a different configuration)

If you need a different setup (e.g., CPU-only, different CUDA version, Linux), you can build the pack on your connected computer.

1. Install uv (on the connected computer):

powershell -ExecutionPolicy Bypass -c "irm https://astral.sh/uv/install.ps1 | iex"

Close and reopen Command Prompt after installing.

2. Create a wrapper project:

mkdir sleap-offline\sleap_offline
cd sleap-offline
type nul > sleap_offline\__init__.py

Create a file called pyproject.toml in the sleap-offline folder with this content:

[project]
name = "sleap-offline"
version = "0.1.0"
requires-python = ">=3.13,<3.14"
dependencies = [
    "sleap[nn]==1.6.1",
]

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[tool.uv]
index-url = "https://download.pytorch.org/whl/cu128"
extra-index-url = ["https://pypi.org/simple"]

Change cu128 to cpu if the offline computer has no NVIDIA GPU. This also makes the pack much smaller (~500 MB vs ~3.4 GB).

3. Lock dependencies and build the pack:

uv lock --python 3.13

set PYTHONIOENCODING=utf-8
uvx uv-pack --verbose --pip-download "--extra-index-url https://download.pytorch.org/whl/cu128"

PowerShell version (click to expand)

uv lock --python 3.13

$env:PYTHONIOENCODING = "utf-8"
uvx uv-pack --verbose --pip-download "--extra-index-url https://download.pytorch.org/whl/cu128"

This produces a pack\ folder. Copy it to a USB drive and follow the same unpack steps from Option A.

---

Important notes

• NVIDIA drivers must be installed on the offline computer. The CUDA libraries come bundled with PyTorch, but the NVIDIA driver itself needs to be installed separately. You can check by running nvidia-smi in Command Prompt. If it's not installed, download the driver installer from https://www.nvidia.com/drivers/ on the connected computer and transfer it via USB.
• You don't need Anaconda for this workflow — everything (including Python) is self-contained in the pack. Anaconda won't interfere though.
• Both computers must be the same platform (e.g., both Windows). You can't pack on Windows and unpack on Linux.

Let us know how it goes!

❤️ Talmo & Claude 🤖

Troubleshooting

"tar: Cannot connect to C: resolve failed" during unpack

This happens if Git for Windows is installed — its tar command conflicts with the Windows built-in tar. Fix by extracting the bundled Python manually first:

mkdir pack\.python
C:\Windows\System32\tar.exe -C pack\.python -xzf pack\python\cpython-3.13.12+20260211-x86_64-pc-windows-msvc-install_only_stripped.tar.gz

Then re-run pack\unpack.cmd.

torch.cuda.is_available() returns False

The NVIDIA driver is missing. Run nvidia-smi — if it's not found, you need to install the NVIDIA driver. Download the installer on the connected computer from https://www.nvidia.com/drivers/ and transfer it.

Zip file won't extract with "Extract All"

The zip is ~3.4 GB. Windows' built-in extractor can sometimes struggle with large files. Use tar in Command Prompt instead:

tar -xf sleap-offline-pack-win-cu128.zip

Or install 7-Zip (can also be transferred offline).

Unicode errors when running uv-pack

Make sure PYTHONIOENCODING=utf-8 is set before running (shown in the commands above).

[Saheli-R]

Thanks @talmo for your detailed instructions! :)

I only got to try your workflow today, hence the delayed reply.

I followed option A and tried installing using command prompt. python and sleap both got installed but running the command sleap doctor gives me OSerror - (winError 1114) dynamic link library initialization routine failed. Error loading c10.dll or one of its dependencies.

I have checked that the c10.dll file does exist, but when i try loading/importing it using the file's location in my computer, it shows the same error again. I am stuck at this point.
If more information helps, I have Nvidia 610.62, cuda 13.3 installed in the offline computer.

Could you perhaps help?

Thank you,
Saheli

[Saheli-R]

Thanks @talmo for your detailed instructions! :)

I only got to try your workflow today, hence the delayed reply.

I followed option A and tried installing using command prompt. python and sleap both got installed but running the command sleap doctor gives me OSerror - (winError 1114) dynamic link library initialization routine failed. Error loading c10.dll or one of its dependencies.

I have checked that the c10.dll file does exist, but when i try loading/importing it using the file's location in my computer, it shows the same error again. I am stuck at this point.
If more information helps, I have Nvidia 610.62, cuda 13.3 installed in the offline computer.

Could you perhaps help?

Thank you,
Saheli