An app which helps to estimate GPU VRAM needed for training/inference transformer runs. Check it out here: vram.asmirnov.xyz.

You can find main logic of calculating the result in ./app/_lib/index.ts.

npm install
npm run dev

npm install
npm run build

It will result with ./out folder which can be served with any webserver (e.g. nginx).

• Quantized models inference
• LoRA/QLoRA fine-tuning