Loading…
Loading…
Models / gemma-4 / gemma-4 31B
Text generation · served with vLLM · ↓ 64.4K downloads a month · View on Hugging Face
ErtasAI/MedErtas-v0.1-31B-LS needs about 38 GB of GPU memory for conversations of up to 8,192 tokens. It runs on 4x RTX 3060 from $0.35/h right now; best value: RTX 4090 at $0.46/h.
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 | Deploy |
|---|---|---|---|---|---|
| Originalgoogle/gemma-4-31B | vLLM | 63 GB | 80.7 GB | $0.38/h Nvidia GB10 | Deploy |
| itgoogle/gemma-4-31B-it | vLLM | 63 GB | 80.7 GB | $0.38/h Nvidia GB10 | Deploy |
| IT NVFP4 (nvidia)nvidia/Gemma-4-31B-IT-NVFP4 | vLLM | 33 GB | 47.4 GB | $0.38/h Nvidia GB10 | Deploy |
| it AWQ 4bit (cyankiwi)cyankiwi/gemma-4-31B-it-AWQ-4bit | vLLM | 21 GB | 34.4 GB | $0.32/h 2x RTX 3080 | Deploy |
| it AWQ (QuantTrio)QuantTrio/gemma-4-31B-it-AWQ | vLLM | 20 GB | 33.9 GB |
$0.46/h from your balance while it runs, billed by the hour; stop it from its page. GridPup rents the machine, which downloads the model from Hugging Face (usually ready in 10 to 15 minutes), and gives you an https address, keys and a Playground.
| Deploy |
| it FP8 block (RedHatAI)RedHatAI/gemma-4-31B-it-FP8-block | vLLM | 33 GB | 48.1 GB | $0.38/h Nvidia GB10 | Deploy |
| it NVFP4 (RedHatAI)RedHatAI/gemma-4-31B-it-NVFP4 | vLLM | 23 GB | 37 GB | $0.32/h 2x RTX 3080 | Deploy |
| it FP8 dynamic (RedHatAI)RedHatAI/gemma-4-31B-it-FP8-dynamic | vLLM | 33 GB | 48.1 GB | $0.38/h Nvidia GB10 | Deploy |
| MedErtas v0.1 31B LS (ErtasAI)ErtasAI/MedErtas-v0.1-31B-LS | vLLM | 24 GB | 38 GB | $0.34/h 4x RTX 3060 |
| it NVFP4 turbo (LilaRest)LilaRest/gemma-4-31B-it-NVFP4-turbo | vLLM | 19 GB | 32.6 GB | $0.32/h 2x RTX 3080 | Deploy |
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 of 90% or more that download at 200 Mbit/s or more are offered: the machine fetches the model itself.