lex-au/Orpheus-3b-Kaya-FP16.gguf

Orpheus-3b-Kaya-FP16

This is a fine-tuned version of the pretrained model canopylabs/orpheus-3b-0.1-pretrained, trained on a custom voice dataset and quantised to GGUF FP16 format for fast, efficient inference.

---

🔧 Model Details

• Model Type: Text-to-Speech (TTS)
• Architecture: Token-to-audio language model
• Parameters: ~3 billion
• Quantisation: 8-bit GGUF (FP16)
• Sampling Rate: 24kHz mono
• Training Epochs: 1
• Training Dataset: lex-au/Orpheus-3b-Kaya
• Languages: English

---

🚀 Quick Usage

This model is designed for use with Orpheus-FastAPI, an OpenAI-compatible inference server for text-to-speech generation.

Compatible Inference Servers

You can load this model into:

• GPUStack
• LM Studio
• llama.cpp
• Any other GGUF-compatible OpenAI-style server

📜 License

Apache License 2.0 — free for research and commercial use.

---

🙌 Credits

• Original model by: Canopy Labs
• Fine-tuned, quantised, and API-wrapped by: Lex-au via Unsloth and Huggingface's TRL library.

---

📚 Citation

@misc{orpheus-tts-2025,
  author = {Canopy Labs},
  title = {Orpheus-3b-0.1-pt: Pretrained Text-to-Speech Model},
  year = {2025},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/canopylabs/orpheus-3b-0.1-pt}}
}

@misc{orpheus-kaya-2025,
  author = {Lex-au},
  title = {Orpheus-3b-Kaya-FP16: Fine-Tuned TTS Model (Quantised)},
  note = {Fine-tuned from canopylabs/orpheus-3b-0.1-pt},
  year = {2025},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/lex-au/Orpheus-3b-Kaya-FP16}}
}