Datasets:
The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
MOSS-TTS-Nano bridge codes (3000 examples)
Greedy audio token codes produced by the frozen OpenMOSS-Team/MOSS-TTS-Nano
for an Ultravox-style speech-to-speech bridge (EuroLLM-1.7B hidden states -> frozen MOSS-TTS-Nano).
Self-distillation targets generated with model.inference(do_sample=False, mode="voice_clone")
(Iasmin reference voice, audio_repetition_penalty=1.5, max_new_frames=375).
Codec = MOSS-Audio-Tokenizer-Nano (16 RVQ codebooks @ 12.5 Hz, audio vocab 1024).
cache.pt is a pickled list[dict] (3000 items). Per item: input_ids/labels (MOSS packed,
last dim 17 = 1 text + 16 audio codebooks), euro_hidden (EuroLLM last-layer hidden, fp16),
align (char-offset MOSS->EuroLLM token map), span/mlen (answer-text injection span).
This file grows as generation continues. Load with torch.load("cache.pt", map_location="cpu").
License: codes derived from MOSS-TTS-Nano (weights lfm1.0 / code Apache-2.0).
- Downloads last month
- 25