gemma-2-2b
by google
2.6B params · text-generation · 661 likes · 235.9k downloads
gemma-2-2b is a 2.6B parameter model. At Q4 quantization it requires 1GB of VRAM. It runs comfortably on GeForce RTX 4090 (502 tok/s), GeForce RTX 5090 (753 tok/s), M4 Max 128GB (183 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Featherless |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 502 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 753 tok/s | comfortable |
| M4 Max 128GB | 128GB | 183 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 91 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 91 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 920 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 1850 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 444 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 367 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 356 tok/s | comfortable |