Professor/kinyarwandatts

LoRA fine-tune of OuteAI/Llama-OuteTTS-1.0-1B for Kinyarwanda text-to-speech.

Model Details

• Developer: Professor
• Base model: OuteAI/Llama-OuteTTS-1.0-1B
• Languages: ['rw']
• License: apache-2.0
• Library: 🤗 Transformers / TRL / Unsloth

How to Use

Install instructions / weights coming soon.

Note: Exact audio synthesis depends on your TTS pipeline/vocoder.

Training Details

• Dataset: mbazaNLP/kinyarwanda-tts-dataset (~4k pairs)
• Trainer: TRL SFTTrainer (Unsloth)
• Hyperparameters:

{
  "note": "See trainer_state.json in training_artifacts."
}

• Hardware:

{
  "machine": "x86_64",
  "system": "Linux",
  "python": "3.12.11",
  "gpu_name": "Tesla T4",
  "gpu_count": "1",
  "gpu_capability": "(7, 5)",
  "cuda": "12.6",
  "bf16_support": "False"
}

• Logs & checkpoints: see training_artifacts/ (TensorBoard events & trainer_state.json).

Evaluation

• Track eval_loss in TensorBoard. Add MOS or intelligibility metrics if available.

Limitations & Risks

• Kinyarwanda-only; may mispronounce OOV words or names.

Citation

@misc{Professor_kinyarwandatts,
  author = {Olufemi Victor Tolulope},
  title  = {Professor/kinyarwandatts: Kinyarwanda TTS LoRA},
  year   = {2025},
  url    = {https://huggingface.co/Professor/kinyarwandatts}
}