Kushtrim

Qwen3-ASR-1.7B-Norwegian-2306

README

License: apache-2.0

Model details

Table with columns: Property, Value
Property	Value
Task	Automatic speech recognition
Language	Norwegian
Base model	Qwen3-ASR-1.7B
Architecture	`Qwen3ASRForConditionalGeneration`
Saved precision	BF16
Audio feature size	128
Saved training step	152,676
Recorded epoch	2.258
Per-device training batch size	16

The saved trainer state contains no evaluation history, so this card does not claim a WER or other benchmark score.

Usage

bash
pip install -U qwen-asr

python
import torch
from qwen_asr import Qwen3ASRModel

model = Qwen3ASRModel.from_pretrained(
    "Kushtrim/Qwen3-ASR-1.7B-Norwegian-2306",
    dtype=torch.bfloat16,
    device_map="cuda:0",
    max_inference_batch_size=32,
    max_new_tokens=256,
)

results = model.transcribe(
    audio="path/to/audio.wav",
    language="Norwegian",
)

print(results[0].text)

For automatic language identification, set language=None.

Intended use

Norwegian audio transcription and ASR research.
Comparison with the base Qwen3-ASR model and other Norwegian checkpoints.
Further fine-tuning or domain adaptation where the base-model license allows.

Limitations

No benchmark results are currently documented for this checkpoint.
Accuracy may vary across Bokmål, Nynorsk, dialects, recording conditions, background noise, and specialized vocabulary.
Generated transcripts can contain omissions, substitutions, or invented words. Human review is recommended for consequential use.
Evaluate the model on representative audio before deployment.

Provenance

The architecture, precision, training step, epoch, and batch size above were recovered from the repository's saved configuration and trainer state. Add the training datasets, preprocessing decisions, and evaluation results here when they are available.

Development and attribution

Original architecture and base weights: Qwen Team, through Qwen/Qwen3-ASR-1.7B
Norwegian adaptation, fine-tuning, training-data curation, checkpoint preparation, packaging, and release: Kushtrim Visoka
Model card and ongoing maintenance: Kushtrim Visoka

Acknowledgements

This work builds on Qwen3-ASR by the Qwen team. The original architecture and base weights remain the work of their respective authors. This repository documents the Norwegian fine-tuning and release work carried out by Kushtrim Visoka and does not claim authorship of the original Qwen3-ASR model.

Citation

If you use this fine-tuned model, please cite this repository and the original Qwen3-ASR model.

bibtex
@misc{visoka_qwen3_asr_17b_norwegian_2306_2026,
  author       = {Visoka, Kushtrim},
  title        = {Qwen3-ASR-1.7B Norwegian 2306},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/Kushtrim/Qwen3-ASR-1.7B-Norwegian-2306},
  note         = {Fine-tuned from Qwen/Qwen3-ASR-1.7B}
}

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

Kushtrim

Model Tree

Base

Qwen/Qwen3-ASR-1.7B

Fine-tuned

this model

Input Modalities

Audio

Output Modalities

Text

Supported Functionality