Loading…
Loading…
Models / gemma-2
3 models (2B to 27B) · ↓ 1.5M downloads a month · from $0.10/h on RTX 3060 right now. Each model's versions do the same job; quantized ones (AWQ, FP8, 4-bit) need less GPU memory and run on cheaper machines.
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|---|---|---|---|
| Originalgoogle/gemma-2-9bLicence to accept | vLLM | 18 GB | ~24.4 GB | $0.34/h 2x RTX 5060 Ti |
| itgoogle/gemma-2-9b-itLicence to accept | vLLM | 18 GB | ~24.4 GB | $0.34/h 2x RTX 5060 Ti |
| it AWQ INT4 (hugging-quants)hugging-quants/gemma-2-9b-it-AWQ-INT4 | vLLM | 6.2 GB | 12.2 GB | $0.14/h RTX A4000 |
| bnb 4bit (unsloth)unsloth/gemma-2-9b-bnb-4bit | vLLM | 6.1 GB | 12.2 GB | $0.14/h RTX A4000 |
| it bnb 4bit (unsloth)unsloth/gemma-2-9b-it-bnb-4bit | vLLM | 6.1 GB | 12.2 GB | $0.14/h RTX A4000 |
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|---|---|---|---|
| Originalgoogle/gemma-2-27bLicence to accept | vLLM | 54 GB | ~67.6 GB | $0.89/h 2x Nvidia A40 |
| itgoogle/gemma-2-27b-itLicence to accept | vLLM | 54 GB | ~67.6 GB | $0.89/h 2x Nvidia A40 |
| datagemma rag 27b itgoogle/datagemma-rag-27b-itLicence to accept | vLLM | 54 GB | ~67.6 GB | $0.89/h 2x Nvidia A40 |
| datagemma rig 27b itgoogle/datagemma-rig-27b-itLicence to accept | vLLM | 54 GB | ~67.6 GB | $0.89/h 2x Nvidia A40 |
Model details come from Hugging Face and are refreshed daily; prices follow the live offers. GPU memory is GridPup's estimate: the weights, room for an 8,192-token conversation (chat models) and the runtime's own memory. Large models are split across the GPUs of one machine ("4x RTX 4090").