bohemond92140's picture
Add the model card
7db8e60 verified
|
Raw History Blame Contribute Delete
1.7 kB
---
license: apache-2.0
base_model:
- OpenMOSS/MOSS-TTS-Realtime
pipeline_tag: text-to-speech
library_name: gguf
tags:
- gguf
- tts
- moss
- speech
- openbloom
- kriskross
- quantized
language:
- en
- zh
---
# MOSS-TTS-Realtime 2.0.0 — GGUF (Q4_K_M)
OpenBloom Labs' GGUF quantization of [MOSS-TTS-Realtime 2.0.0](https://huggingface.co/OpenMOSS/MOSS-TTS-Realtime)
— a realtime streaming text-to-speech model from the OpenMOSS team.
## Files
| File | Size | sha256 |
|---|---|---|
| `moss-tts-realtime-q4_k_m.gguf` | 1,335,751,072 B (~1.24 GiB) | `ccd2ab76524affbc3a33a54f56224e24eabb008e73dd2b4142e8bf48cf492ed6` |
The quantization keeps the local codebook embeddings at f16 (the host
lookups require f16/f32 rows); everything else rides Q4_K_M.
## Usage
The conversion targets llama.cpp via **[KrisKross](https://huggingface.co/OpenBloom/runtime-releases)**
— OpenBloom's build of the OpenMOSS llama.cpp fork (`llama-moss-tts` +
`llama-moss-tts-realtime`, the streaming duplex lane). Any
llama.cpp-compatible loader that carries the MOSS-TTS architecture can
serve it too.
```
# verify before use
sha256sum moss-tts-realtime-q4_k_m.gguf
ccd2ab76524affbc3a33a54f56224e24eabb008e73dd2b4142e8bf48cf492ed6
```
## License
The MOSS-TTS model is **Apache-2.0** (© the OpenMOSS team and
contributors); this quantization inherits the license verbatim.
## Attribution
The model, the architecture and the training belong to
[OpenMOSS](https://github.com/OpenMOSS/MOSS-TTS). OpenBloom Labs
produced the GGUF conversion and publishes it for the OpenBloom
installer's local TTS lanes (one-shot synthesis + the realtime duplex
voice loop). Not affiliated with or endorsed by the OpenMOSS team.