Beta

Explore
Marketplace
Neural Labs
Playground
Wallet
Docs

Beta

Explore Marketplace Neural Labs Wallet Docs

Build

Manage

My assets Manage wallet Organization

Community

Discord X (Twitter)Telegram

Docs Privacy Policy Terms and Conditions

Proud member of

Proud member of

Docs Cookie Policy

Privacy Policy Terms and Conditions

© 2026 AlphaNeural AI. All rights reserved.

granite-speech-4.1-2b-nar-mlx-5bit – AI Model by mlx-community | AlphaNeural AI

You can deploy this model and start earning money today!

mlx-community/
granite-speech-4.1-2b-nar-mlx-5bit

like

0

granite_speech_nar

non-autoregressive

automatic-speech-recognition

ibm-granite/granite-speech-4.1-2b-nar

Views

No views yet

Granite Speech 4.1 2B NAR — MLX

MLX port of ibm-granite/granite-speech-4.1-2b-nar for Apple Silicon. Runs via mlx-audio.

Architecture

Non-autoregressive ASR via CTC + bidirectional LM editing:

16-layer Conformer encoder (543M params) produces an initial BPE CTC hypothesis.
2-layer windowed Q-Former projector (80M params) converts multi-layer encoder states into audio embeddings.
40-layer bidirectional Granite editor (1.6B params) takes [audio | hypothesis_tokens] and emits edited logits in a single forward pass — no autoregression, no KV cache.
Final CTC collapse on text-position logits yields the transcript.

Total: ~2.25B params, bf16.

Quickstart

python

from pathlib import Path
from mlx_audio.stt.utils import load_model

model = load_model(Path("mlx-community/granite-speech-4.1-2b-nar-mlx"))
out = model.generate("audio.wav")
print(out.text)

Limitations

Batch size 1.
bf16 baseline only — no quantized variants yet.
No streaming inference.
macOS 14+, Apple Silicon.

Reference

Upstream model card: https://huggingface.co/ibm-granite/granite-speech-4.1-2b-nar

Validated against the upstream PyTorch reference: exact 44-token match and exact transcript string on the example wav.

License

Apache-2.0, matching the upstream model.