Loading…
Loading…
Models / gemma-4-E2B / gemma-4-E2B ~5.12B
Any to any · served with ONNX Runtime · ↓ 13.9K downloads a month · View on Hugging Face
onnx-community/gemma-4-E4B-it-ONNX needs about 2.8 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.
Smaller versions (8-bit, 4-bit) need less GPU memory and answer a little less precisely.
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 |
|---|---|---|---|---|
| Originalgoogle/gemma-4-E2B | vLLM | 10 GB | 13.9 GB | $0.13/h RTX 4060 Ti |
| itgoogle/gemma-4-E2B-it | vLLM | 10 GB | 13.9 GB | $0.13/h RTX 4060 Ti |
| it qat w4a16 ctgoogle/gemma-4-E2B-it-qat-w4a16-ct | vLLM | 8.3 GB | 11.8 GB | $0.09/h RTX 3060 |
| it qat q4 0 unquantizedgoogle/gemma-4-E2B-it-qat-q4_0-unquantized | vLLM | 10 GB | 13.9 GB | $0.13/h RTX 4060 Ti |
| it qat mobile ctgoogle/gemma-4-E2B-it-qat-mobile-ct | vLLM | 2.7 GB | 5.5 GB | $0.09/h RTX 3060 |
| it unsloth bnb 4bit (unsloth)unsloth/gemma-4-E2B-it-unsloth-bnb-4bit | vLLM | 8.1 GB | 11.6 GB | $0.09/h RTX 3060 |
Deploying ONNX models from the catalog is coming soon. Today, download the files from Hugging Face and upload them in Inference.
| it AWQ INT4 (cyankiwi)cyankiwi/gemma-4-E2B-it-AWQ-INT4 | vLLM | 7.6 GB | 11 GB | $0.09/h RTX 3060 |
| gemma 4 E4B it ONNX (onnx-community)onnx-community/gemma-4-E4B-it-ONNX | ONNX Runtime | 100 MB | 1.2 GB | $0.07/h Quadro P4000 |
| it ONNX (onnx-community)onnx-community/gemma-4-E2B-it-ONNX | ONNX Runtime | 100 MB | 1.1 GB | $0.07/h Quadro P4000 |
| it qat mobile ONNX (onnx-community)onnx-community/gemma-4-E2B-it-qat-mobile-ONNX | ONNX Runtime | 90 MB | 1.1 GB | $0.07/h Quadro P4000 |
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.