Datasetpublic

stokiz/moss-ttsd-sglang-assets

MOSS-TTSD SGLang audio-tokenizer assets payload This private Kaggle dataset is generated by Phorcys.Tools.MossTtsdAssetsUploader for PHRunner.Kaggle.Service.MossTtsGgufD. Purpose: fully offline fallback decoding of SGLang /generate responses that return output_ids instead of a WAV payload. Repository-visible payload: MOSS-Audio-Tokenizer-ONNX__encoder.onnx, MOSS-Audio-Tokenizer-ONNX__encoder.data, MOSS-Audio-Tokenizer-ONNX__decoder.onnx, MOSS-Audio-Tokenizer-ONNX__decoder.data… See the full description on the dataset page: https://huggingface.co/datasets/stokiz/moss-ttsd-sglang-assets.

sourceHugging Faceupdated 18d agoView on Hugging Face

0likes165downloads

Dataset Card

MOSS-TTSD SGLang audio-tokenizer assets payload

This private Kaggle dataset is generated by Phorcys.Tools.MossTtsdAssetsUploader for PHRunner.Kaggle.Service.MossTtsGgufD.

Purpose: fully offline fallback decoding of SGLang /generate responses that return output_ids instead of a WAV payload.

Repository-visible payload:

—MOSS-Audio-Tokenizer-ONNX__encoder.onnx, MOSS-Audio-Tokenizer-ONNX__encoder.data, MOSS-Audio-Tokenizer-ONNX__decoder.onnx, MOSS-Audio-Tokenizer-ONNX__decoder.data, and companion top-level files from niobures/MOSS-Audio-Tokenizer-ONNX.

The SGLang service kernel mounts this dataset and recreates MOSS-Audio-Tokenizer-ONNX/ from assets_manifest.json using symlinks under /tmp/phorcys_moss_work. This avoids Hugging Face downloads and avoids large archive extraction during production runs.

—assets_manifest.json, moss_ttsd_sglang_assets_manifest.json, moss_ttsd_audio_tokenizer_manifest.json: source paths, sizes, and SHA-256 checksums.

—moss-audio-tokenizer-onnx: 10 files.

Configure the service with:

text

MOSS_TTSD_AUDIO_TOKENIZER_DATASET_REF=kaggle-pool-account/moss-ttsd-sglang-assets
MOSS_TTSD_ALLOW_HF_AUDIO_TOKENIZER_DOWNLOAD=false