Loading…
Loading…
Models / whisper-large / whisper-large ~1.54B
Speech to text · served with ONNX Runtime · ↓ 15.4K downloads a month · View on Hugging Face
xkeyC/whisper-large-v3-turbo-gguf needs about 3.6 GB of GPU memory. It runs on one Quadro P4000 from $0.07/h right now; best value: RTX 5070 Ti at $0.19/h.
These ONNX files are an export for running in a browser (an audio encoder and a text decoder, each a version here): served as tensors, they do not make a transcript. For speech to text, pick the original model (for example openai/whisper-base), served with vLLM.
The same model packaged differently: a quantized version needs less GPU memory and runs on a cheaper machine.
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|---|---|---|---|
| Originalopenai/whisper-large-v3 | vLLM | 9.3 GB | 13.1 GB | $0.13/h RTX A4000 |
| turboopenai/whisper-large-v3-turbo | vLLM | 1.6 GB | 3.9 GB | $0.09/h RTX 3060 |
| FP8 dynamic (RedHatAI)RedHatAI/whisper-large-v3-FP8-dynamic | vLLM | 1.7 GB | 4 GB | $0.09/h RTX 3060 |
| kb whisper large (KBLab)KBLab/kb-whisper-large | vLLM | 3.2 GB | 5.9 GB | $0.09/h RTX 3060 |
| turbo (onnx-community)onnx-community/whisper-large-v3-turbo | ONNX Runtime | 190 MB | 1.3 GB | $0.07/h Quadro P4000 |
| turbo timestamped (onnx-community)onnx-community/whisper-large-v3-turbo_timestamped | ONNX Runtime | 190 MB | 1.3 GB |
Deploying ONNX models from the catalog is coming soon. Today, download the files from Hugging Face and upload them in Inference.
| turbo gguf (xkeyC)xkeyC/whisper-large-v3-turbo-gguf | ONNX Runtime | 1.8 GB | 3.6 GB | $0.07/h Quadro P4000 |
| turbo (TheStageAI)TheStageAI/thewhisper-large-v3-turbo | vLLM | 1.6 GB | 3.9 GB | $0.09/h RTX 3060 |
Prices are per hour for the whole machine and its disk, right now. Speed is Vast's DLPerf score of the machine, relative to the cheapest. Best value: the most speed per dollar. Only machines with a Vast reliability above 95% that download at 400 Mbit/s or more are offered: the machine fetches the model itself.