How much GPU memory does moonshine-base-ONNX need?

About 1.2 GB: 0.17 GB of weights and the runtime's own memory.

What is the cheapest GPU to run moonshine-base-ONNX?

Right now one Quadro P4000 at $0.07 an hour; RTX 5090 at $0.67 an hour is the best value for its speed. Prices are live offers from GPU clouds, checked every 15 minutes.

Can moonshine-base-ONNX run on an RTX 4090?

Yes: it needs about 1.2 GB and an RTX 4090 has 24 GB.

Models / moonshine

onnx-community/moonshine-base-ONNX

Speech to text · served with ONNX Runtime · ↓ 21.7K downloads a month · View on Hugging Face

onnx-community/moonshine-base-ONNX needs about 1.2 GB of GPU memory. It runs on one Quadro P4000 from $0.07/h right now; best value: RTX 5090 at $0.67/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.

Weights: 170 MB
GPU memory needed: about 1.2 GB
Memory · disk: 8 GB · 30 GB
Licence: mit

Versions

Smaller versions (8-bit, 4-bit) need less GPU memory and answer a little less precisely.

full precisiononnx/decoder_model.onnx170 MB · 1.2 GB GPU 4-bit (bnb)onnx/decoder_model_bnb4.onnx70 MB · 1.1 GB GPU float16onnx/decoder_model_fp16.onnx80 MB · 1.1 GB GPU 8-bitonnx/decoder_model_int8.onnx40 MB · 1.1 GB GPU full precisiononnx/decoder_model_merged.onnx170 MB · 1.2 GB GPU 4-bit (bnb)onnx/decoder_model_merged_bnb4.onnx70 MB · 1.1 GB GPU float16onnx/decoder_model_merged_fp16.onnx160 MB · 1.2 GB GPU 8-bitonnx/decoder_model_merged_int8.onnx

Hardware and deploy

Deploying ONNX models from the catalog is coming soon. Today, download the files from Hugging Face and upload them in Inference.

Questions

How much GPU memory does moonshine-base-ONNX need?: About 1.2 GB: 0.17 GB of weights and the runtime's own memory.
What is the cheapest GPU to run moonshine-base-ONNX?: Right now one Quadro P4000 at $0.07 an hour; RTX 5090 at $0.67 an hour is the best value for its speed. Prices are live offers from GPU clouds, checked every 15 minutes.
Can moonshine-base-ONNX run on an RTX 4090?: Yes: it needs about 1.2 GB and an RTX 4090 has 24 GB.

Loading…