gemma-3n-E4B-it
by google
7.8B params · image-text-to-text · 918 likes · 21.0k downloads
gemma-3n-E4B-it is a 7.8B parameter model. At Q4 quantization it requires 4GB of VRAM. It runs comfortably on GeForce RTX 4090 (167 tok/s), GeForce RTX 5090 (250 tok/s), M4 Max 128GB (61 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Together AI | 54 tok/s |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 167 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 250 tok/s | comfortable |
| M4 Max 128GB | 128GB | 61 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 30 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 30 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 306 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 616 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 147 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 122 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 118 tok/s | comfortable |