Navigate
Home How it works
What's included Pricing FAQ
Back to home
Two ways to get started.
Buy the bundle once, or subscribe and stay current as Colab evolves. Both include all wheels, a personal token, and instant pip install — zero collisions with the default Colab stack.
MissingLink Pro

Pro Subscription

Stay current as Colab updates

$9 / month
7-day free trial — no charge until day 8
  • Everything in the Survival Pack
  • New wheels as packages update
  • Rebuilt when Colab runtime changes
  • Always frictionless — tested against the latest default stack
  • Priority access to new notebooks
  • Token-based auth — no login
  • Cancel anytime, instantly
Start free trial

Billed monthly by Stripe. We never see your card. Cancel or manage anytime.

Which should I pick?

If you just need the current set of wheels and don't care about future updates, the Colab Survival Pack is the best value — one payment and you're set. If you want to stay current every time Google updates the Colab runtime or when new package versions drop, the Pro Subscription keeps you covered automatically.

Just need one wheel?

Buy individual packages if you only need a specific wheel.

Flash Attention
flash-attn 2.7.3 + 2.8.3 Flash Attention 2 — both versions included. Guaranteed for A100, L4 & T4 on Colab.
$5

More individual wheels coming soon. Need a specific one? Support

Complete package list

Both plans include every wheel below, prebuilt and guaranteed to work against CUDA 12.8, PyTorch 2.10, and Python 3.12 on Google Colab. Every wheel installs cleanly alongside the default Colab stack — no uninstalls, no version conflicts, no battling dependencies.

flash-attn 2.8.3

Flash Attention 2 — efficient transformer inference on A100/L4/T4.

🔧

xformers 0.0.35

Memory-efficient attention and transformer building blocks from Meta.

🔺

nvdiffrast 0.4.0

NVIDIA's differentiable rasterizer for 3D deep learning.

🧊

cumesh 0.0.1

CUDA-accelerated mesh processing for 3D pipelines.

🧮

flex-gemm 0.0.1

Flexible GEMM kernels for mixed-precision on Ampere GPUs.

📦

o-voxel 0.0.1

CUDA voxel utilities for 3D pipeline work.

🎨

nvdiffrec-render 0.0.0

Neural rendering components for reconstruction pipelines.

📐

utils3d 0.0.2

3D math and geometry with CUDA acceleration.

🖼️

stable_diffusion_cpp

Run GGUF-quantized Stable Diffusion models. Compiled for frictionless T4 compatibility.

Guaranteed environment

SpecDetails
GPUNVIDIA A100 NVIDIA L4 NVIDIA T4
PlatformGoogle Colab linux x86_64
Python3.12
CUDA12.8
PyTorch2.10