Instructions to use rumik-ai/rumik-oss-1-mlx-8bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use rumik-ai/rumik-oss-1-mlx-8bit with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir rumik-oss-1-mlx-8bit rumik-ai/rumik-oss-1-mlx-8bit
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
rumik-oss 1 · MLX 8-bit
8-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.
Recommended. Near-lossless: 96% next-token agreement with the bf16 reference.
this export
| quantization | 8-bit affine, group size 64, stop predictor kept in bf16 |
| size on disk | 3.6 GB |
| peak memory | 3.8 GB |
| decode speed, M5 MacBook Air (16 GB) | 33-40 tokens/s (0.33-0.40x real time) |
| next-token agreement with bf16 | 96% |
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-8bit")
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-8bit \
--text "नमस्ते, आज आपका दिन कैसा रहा?" --voice Ira \
--instruct "happy, Hindi accent, steady pace" --stream
controls
| control | values |
|---|---|
| voice | Ira, Aisha, Siya, Zoya |
| tone | happy, sad, angry, excited, professional |
| accent | Hindi, Telugu, Tamil, Kannada, Bengali, Punjabi, Indian English |
| pace | slow, fast, steady |
| inline | <laugh>, <chuckle>, <sigh> |
The description can also be written inline: <description="excited, Hindi accent, fast pace"> जल्दी आओ! <laugh>
samples
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 8-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@unpublished{govindu2026rumikoss1,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}
}
- Downloads last month
- 17
8-bit
Model tree for rumik-ai/rumik-oss-1-mlx-8bit
Base model
CohereLabs/tiny-aya-base