OPENMOSS / MULTILINGUAL TTS / 2026

MOSS-TTS-Nano-100M

TOTAL100Mparameters

ACTIVE100Mruntime path

DEPTH12 + localrepeated blocks

HIDDEN768main trunk

CONTEXT32Kstream / sequence

CHECKPOINT706.97MB repo1 shards · Apache-2.0

INTERACTIVE SYSTEM BLUEPRINT

MODEL MRI

128 tokens → 10 seconds · 78.31G arithmetic FLOPs · click any module for evidence

VERIFIED ADAPTER

INPUT128TOKENS

TRANSFORMER STACK12 GLOBAL + 1 LOCALGPT-style global model + local codebook transformer

Pre-Norm Residual

OUTPUT10SECONDS

wraps every repeated block

GLOBAL-LOCAL ATTENTION · PURE AUTOREGRESSIVE AUDIO没有大型 DiT；Audio Tokenizer 与小型 LLM 组成可解释的纯自回归链路。

Audio Tokenizer Nano离散 tokenizer 将高采样率立体声音频压缩为 16 路 code。

GLOBAL12L768

LOCAL1Lcodebooks

CODES161,024 each

AUDIO48kHzstereo

LANG20supported

PARAMS100Mdeclared