Langtrain Logo
Langtrain Docs
GuidesAPI referenceSDKs
GitHubOpen the app

Search documentation

Jump to any docs page

  • What is Langtrain
  • Quick start
  • Installation
GitHub

Installation

Install the Python SDK to fine-tune on your own GPU, or use an API key to train in the Langtrain cloud.

  • Python 3.8+
  • Linux or WSL
  • NVIDIA GPU

What you need

To train on your own machine:
  • –Linux, or Windows with WSL
  • –An NVIDIA GPU with a recent CUDA driver. With 4-bit QLoRA, 16 GB of GPU memory is enough for 7B and 8B models; 24 GB gives you room for longer examples
  • –Python 3.8 or later
No GPU? Train in the cloud from the dashboard or the API instead; nothing needs installing.

Install

Use a virtual environment. langtrain-ai[train] brings in PyTorch, Transformers, PEFT, TRL, bitsandbytes and datasets; langtune is the training library underneath.
python
1python -m venv .venv
2source .venv/bin/activate
3

4pip install "langtrain-ai[train]" langtune

Vision models

Add the vision extra to fine-tune vision-language models with Langvision.
python
1pip install "langtrain-ai[vision]"

Check it works

Confirm the package imports and PyTorch can see your GPU.
python
1python -c "import langtrain; print('langtrain', langtrain.__version__)"
2python -c "import torch; print('CUDA', torch.cuda.is_available(), torch.cuda.get_device_name(0) if torch.cuda.is_available() else '')"

API key (cloud only)

Local training needs no account. To start runs in the Langtrain cloud through the REST API, create a key in the dashboard under API keys. Leave it unset in the shell you train locally from: when it is set, Langtune sends runs to the cloud instead.
python
1export LANGTRAIN_API_KEY=sk-lt-...

Next

Train your first model with the Quick start.
PreviousQuick startNextOverview