Jan-nano
by Menlo
4.0B params · text-generation · 502 likes · 3.7k downloads
Jan-nano is a 4.0B parameter model. At Q4 quantization it requires 2GB of VRAM. It runs comfortably on GeForce RTX 4090 (326 tok/s), GeForce RTX 5090 (489 tok/s), M4 Max 128GB (119 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Featherless |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 326 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 489 tok/s | comfortable |
| M4 Max 128GB | 128GB | 119 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 59 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 59 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 598 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 1202 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 288 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 238 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 231 tok/s | comfortable |