Cohere Transcribe 2B — MLX FP16
MLX weights for Cohere Transcribe 2B, ready for local speech-to-text with speech-swift.
| Parameters | Precision | Sample rate | Languages |
|---|---|---|---|
| 2B | FP16 | 16 kHz | 14 |
Files
| File | Description |
|---|---|
model*.safetensors | MLX model weights |
config.json | Model and runtime configuration |
tokenizer.model and tokenizer JSON files | SentencePiece tokenizer data |
Usage
speech transcribe recording.wav --engine cohere --model fp16
import CohereTranscribeASR
let model = try await CohereTranscribeModel.load(
"aufklarer/Cohere-Transcribe-2B-MLX-FP16"
)
let text = model.transcribe(
audio: samples,
sampleRate: sourceSampleRate,
language: "en"
)
Benchmark
English FLEURS read speech, 80 utterances (759.56 seconds), Apple M5 Pro:
| Variant | Bundle | WER | Mean RTF | Peak RSS |
|---|---|---|---|---|
| FP16 | 3.85 GiB | 6.178% | 0.0444 | 3,964 MiB |
| INT5 | 1.62 GiB | 6.288% | 0.0150 | 1,776 MiB |
| INT8 | 2.25 GiB | 6.178% | 0.0159 | 2,428 MiB |
These read-speech measurements are not Artificial Analysis Conversational or ElevenLabs Scribe results.
Source and license
The base model is CohereLabs/cohere-transcribe-03-2026 and is released under Apache 2.0.