gemma-2b-it
by google
2.0B params · text-generation · 860 likes · 59.8k downloads
gemma-2b-it is a 2.0B parameter model. At Q4 quantization it requires 1GB of VRAM. It runs comfortably on GeForce RTX 4090 (656 tok/s), GeForce RTX 5090 (984 tok/s), M4 Max 128GB (240 tok/s).
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 656 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 984 tok/s | comfortable |
| M4 Max 128GB | 128GB | 240 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 120 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 120 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 1202 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 2419 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 580 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 480 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 465 tok/s | comfortable |