int8 AR ONNX (dynamic QInt8, ~1.5x faster CPU, weights-only)

AR ONNX (prefill/decode/local + manifest) for the ORT demo backend

demo: prompt-config manifest (token ids + templates)

demo: codec ONNX (encode+decode for voice clone + streaming)

demo: tokenizer config.json

demo: tokenizer tokenizer_config.json

demo: tokenizer special_tokens_map.json

demo: tokenizer tokenization_moss_tts_nano.py

demo: tokenizer tokenizer.model

split moss_codec.gguf

split moss_nano.gguf

MOSS-TTS-Nano Q8_0 merged AR+codec (verified, RTF 0.35 w/ custom kernel)

card

initial commit