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 devnpm install
npm run buildIt will result with ./out folder which can be served with any webserver (e.g. nginx).
- Quantized models inference
- LoRA/QLoRA fine-tuning