rumik-oss-1-mlx-4bit

<p align="center"> <img src="https://huggingface.co/rumik-ai/rumik-oss-1/resolve/main/assets/rumik-logo.png" alt="rumik.ai" width="100%"> </p> <p align="center"> <a href="https://huggingface.co/rumik-ai/rumik-oss-1"><img src="https://img.shields.io/badge/original_model-rumik--oss_1-555555" alt="original model"></a> <a href="https://rumik.ai/research/rumik-oss"><img src="https://img.shields.io/badge/blog-overview-555555" alt="blog"></a> <a href="https://github.com/Blaizzy/mlx-audio"><img src="https://img.shields.io/badge/runs_with-mlx--audio-000000" alt="mlx-audio"></a> </p>

rumik-oss 1 · MLX 4-bit

4-bit MLX quantization of rumik-oss 1, a 3B multilingual text-to-speech model for 22 Indic languages + English with expressive delivery control and inline vocalizations, producing 24 kHz audio. Runs on Apple silicon with mlx-audio.

Fastest. Noticeable fidelity loss (80% next-token agreement); treat as experimental.

this export

100 audio tokens make one second of speech. The Mimi codec is downloaded automatically on first use.

usage

pip install mlx-audio

from mlx_audio.tts.utils import load

model = load("rumik-ai/rumik-oss-1-mlx-4bit")
for chunk in model.generate(
    "नमस्ते, आज आपका दिन कैसा रहा?",
    voice="Ira",
    instruct="happy, Hindi accent, steady pace",
    stream=True,
):
    play(chunk.audio)  # 24 kHz float32

python -m mlx_audio.tts.generate --model rumik-ai/rumik-oss-1-mlx-4bit \
  --text "नमस्ते, आज आपका दिन कैसा रहा?" --voice Ira \
  --instruct "happy, Hindi accent, steady pace" --stream

controls

The description can also be written inline: <description="excited, Hindi accent, fast pace"> जल्दी आओ! <laugh>

samples

<audio controls preload="none" src="https://huggingface.co/rumik-ai/rumik-oss-1/resolve/main/samples/ira.wav"></audio> <audio controls preload="none" src="https://huggingface.co/rumik-ai/rumik-oss-1/resolve/main/samples/aisha.wav"></audio> <audio controls preload="none" src="https://huggingface.co/rumik-ai/rumik-oss-1/resolve/main/samples/siya.wav"></audio> <audio controls preload="none" src="https://huggingface.co/rumik-ai/rumik-oss-1/resolve/main/samples/zoya.wav"></audio>

Samples above are from the original bf16 model; see the original card for benchmarks (IndicEmo, NoVA, WER/CER) and limitations.

license

Research and non-commercial use under tiny aya fire's CC-BY-NC 4.0 with acceptable-use addendum, unchanged from the original. LICENSE and NOTICE are included; this repo adds 4-bit weight quantization as a modification. The Mimi codec is CC-BY-4.0.

citation

@unpublished{govindu2026rumikoss1,
  title  = {{rumik-oss 1 technical report}},
  author = {Govindu Pranav and Anant Shukla and Suryansh Shakya and Aman Anand and Vatsal Bharti},
  year   = {2026},
  note   = {In preparation}
}