Bielik-1.5B-v3.0-Instruct-onnx-q4

Bielik-1.5B-v3.0-Instruct — ONNX q4

A 4-bit (q4) ONNX re-quantization of speakleash/Bielik-1.5B-v3.0-Instruct, packaged as a single self-contained onnx/model_q4.onnx file (≈1.1 GiB, no external data shard) so it can be loaded directly in the browser via 🤗 transformers.js.

This repository contains only a different serialization/quantization of the original model's weights. No additional training or fine-tuning was performed. All model capabilities, behaviour, and limitations are those of the upstream base model.

Attribution & license

• Base model: speakleash/Bielik-1.5B-v3.0-Instruct
• Original authors: SpeakLeash Team & ACK Cyfronet AGH
• License: Apache-2.0 (inherited from the base model)

Please cite the upstream technical report when using this model:

@misc{ociepa2025bielikv3,
  title         = {Bielik v3 Small: Technical Report},
  author        = {Krzysztof Ociepa and Łukasz Flis and Remigiusz Kinas and
                   Krzysztof Gwoździej and Adrian Wróbel},
  year          = {2025},
  eprint        = {2505.02550},
  archivePrefix = {arXiv}
}

Usage (transformers.js)

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

const generator = await pipeline(
  'text-generation',
  'dx0ne/Bielik-1.5B-v3.0-Instruct-onnx-q4',
  { dtype: 'q4' }
);

Files

config.json
generation_config.json
tokenizer.json
tokenizer_config.json
special_tokens_map.json
onnx/
  model_q4.onnx          # ~1.1 GiB, self-contained