gemma-7b
by google
7.0B params · text-generation · 3.3k likes · 109.9k downloads
gemma-7b is a 7.0B parameter model. At Q4 quantization it requires 4GB of VRAM. It runs comfortably on GeForce RTX 4090 (187 tok/s), GeForce RTX 5090 (281 tok/s), M4 Max 128GB (68 tok/s).
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 187 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 281 tok/s | comfortable |
| M4 Max 128GB | 128GB | 68 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 34 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 34 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 343 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 691 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 165 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 137 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 133 tok/s | comfortable |