This model requires you to accept the MIT license terms before download.
Description
Whisper Large-V3 Turbo — OpenAI's speed-optimized speech-to-text model, a fine-tuned variant of Whisper large-v3 with a pruned decoder (4 layers) that delivers much faster multilingual transcription at near-identical accuracy. MIT license.
Tags
At a glance
| Kind | speech-to-text model |
|---|---|
| Precision | float16 |
| Container size | 1.5 GiB |
| Weights inside | 1.5 GiB |
| Components | 2 |
| Licence | MIT |
What is inside the container
A .nbx container packs every part of whisper-large-v3-turbo — 2 components with weights, graphs and runtime settings — so one file is all the runtime needs.
| Component | Blocks | Layers | Weights |
|---|---|---|---|
| model.decoder | 0 | 4 | 328 MiB |
| model.encoder | 0 | — | 1.2 GiB |
Default settings
The values the container ships with. They apply when you do not pass your own.
- Temperature
- 0
Memory needed
The weights of whisper-large-v3-turbo take 1.5 GiB in float16. That is the floor for the memory the runtime has to place, spread over the devices it finds; activations come on top and grow with the size and length you ask for. A measured peak per device class is not published for this model yet.
Verify what you downloaded
Every download carries its SHA-256, and the file endpoint sends it in the X-NBX-SHA256 header. The container is 1.5 GiB (1,619,218,207 bytes).
$ sha256sum whisper-large-v3-turbo.nbx
99cf9f4b64fd9e70fbb190b2f554330adfd45e00ad8de8f5a9a442aab01ad0ab
Questions
What is whisper-large-v3-turbo?
A speech-to-text model by openai, distributed as a float16 .nbx container and licensed MIT.
How big is the download?
1.5 GiB, of which 1.5 GiB are weights.
How do I run it?
Install the runtime from the download page, then use the commands under “CLI Usage” below: neurobrix import openai/whisper-large-v3-turbo, then neurobrix serve --model whisper-large-v3-turbo.
CLI Usage
$ neurobrix import openai/whisper-large-v3-turbo
$ neurobrix serve --model whisper-large-v3-turbo
$ neurobrix run --prompt "your prompt here"
$ neurobrix stop