Orukeet (MLX)

An MLX conversion of oruk/orukeet at revision b59c13a733fce5cf193230d0fa317ee7f145a108, for mlx-audio-swift's Parakeet loader.

Orukeet is by Oruk AI (Nathan Roll, Irene Yi, Büşra Marşan, Vianney Grenez, Gabriel Stein, Momcilo Mrkaic, Pavle Padjin, Vladimir Zeljkovic, Calbert Graham), fine-tuned from NVIDIA Parakeet TDT 0.6B v3. See the original model card for benchmarks, citation and limitations.

Changes

Format only, no retraining: tensors renamed to the MLX Parakeet layout, convolution weights moved to channels-last, the LSTM's input and hidden biases summed, and weights stored as float16. The NeMo config and tokenizer files come from mlx-community/parakeet-tdt-0.6b-v3 (ed2b7e8c15f9aaa0b5772e2efb986255eaef7e15), which has the same architecture and tokenizer.

License

The weights are licensed under CC-BY-SA-4.0, like the original; this conversion is shared under the same license. The original LICENSE, LICENSE-WEIGHTS and NOTICE.md are included.

Downloads last month
15
Safetensors
Model size
0.6B params
Tensor type
F16
·
MLX
Hardware compatibility
Log In to add your hardware

Quantized

Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for itsreidar/orukeet-mlx

Finetuned
oruk/orukeet
Finetuned
(1)
this model