About
Infrastructure formodels you own.
Langtrain helps developers fine-tune open models on their own data and run them on their own terms, including fully offline.
- Open-source packages4langtrain-ai · langtune · langvision · plimai
- Ways to train2Your own GPU (Langtune) · Langtrain GPUs
Our story
Built for builders. Fine-tuning in a few lines.
Developers were spending more time on CUDA versions, training scripts and serving than on their product. We started Langtrain so that fine-tuning an open model is a few function calls, and the result is yours to run anywhere.
from langtune import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained("meta-llama/Llama-3.1-8B")
model = FastLanguageModel.get_peft_model(model, r=16, method="qlora")
FastLanguageModel.train(model, tokenizer, dataset, output_dir="./output")
FastLanguageModel.save_pretrained(model, tokenizer, "./my-assistant", merge_before_save=True)What drives us
Four principles. Every decision goes through them.
You own what you train
Weights are standard files you keep: on your disk after local training, or in your own Hugging Face repo after a cloud run.
Open by default
Langtune and Langvision are MIT licensed. Read the code, change it and use it commercially.
No ML team required
Sensible defaults and a dataset check that suggests a model and setup, so a developer can train a useful model.
Say what works today
We document features that exist and label the ones that don't, so what you read is what you can run.
The people
Built by a small team that ships.
Pritesh Raj
Co-founder, CEO & CTO
Leads product and engineering at Langtrain, and wrote the first versions of Langtune and Langvision.
Help us build it.We're hiring.
See the open roles, or talk to us about what you're building.