Datasetpublic

Transl/qwen3-omni-open-source-balanced-1200

Qwen3-Omni 30A3 open-source balanced subset This dataset contains 1,200 samples selected from public-source-labelled portions of the Qwen3-Omni 30A3 posttrain recipe. The 8 topics are balanced at 150 samples each. Every item includes topic, public_dataset, public_dataset_confidence, source_id, and provenance_json fields. public_dataset is the canonical per-item public-dataset label. Loading The data/train-*.jsonl shards are ordinary Hugging Face JSONL data files… See the full description on the dataset page: https://huggingface.co/datasets/Transl/qwen3-omni-open-source-balanced-1200.

sourceHugging Faceotherupdated 26d agoView on Hugging Face

0likes55downloads

Dataset Card

Qwen3-Omni 30A3 open-source balanced subset

This dataset contains 1,200 samples selected from public-source-labelled portions of the Qwen3-Omni 30A3 posttrain recipe. The 8 topics are balanced at 150 samples each. Every item includes topic, public_dataset, public_dataset_confidence, source_id, and provenance_json fields. public_dataset is the canonical per-item public-dataset label.

Loading

The data/train-*.jsonl shards are ordinary Hugging Face JSONL data files and can be loaded with:

python

from datasets import load_dataset
ds = load_dataset("Transl/qwen3-omni-open-source-balanced-1200", data_files="data/train-*.jsonl", split="train")

The model-ready fields are input_ids, labels, loss_masks, attention_mask, and mm_token_type_ids. Media fields retain environment-local paths and are not copied into this Hub dataset. See export_summary.json for topic counts, source counts, recipe commit, and licensing/provenance notes.