import { pipeline } from '@huggingface/transformers';

const generator = await pipeline(
  'text-generation',
  'bnicenboim/gpt2-medium-dutch-onnx'
);

const result = await generator('De kat zat op de', { max_new_tokens: 50 });
console.log(result[0].generated_text);

Model details

Base model: yhavinga/gpt2-medium-dutch
Architecture: GPT-2 medium (~355M parameters)
Language: Dutch (nl)
Format: ONNX (fp32) + quantized int8
Exported with: 🤗 Optimum
Training corpus: Dutch mC4 corpus (cleaned and deduplicated)

Intended use

Causal language modelling / text completion in Dutch, running client-side in the browser via Transformers.js. Not an instruction or chat model.

Performance note

This is a 355M parameter model. For faster inference in the browser, consider using bnicenboim/gpt-neo-125m-dutch-onnx instead (125M parameters).