Phi-4-mini-instruct
by microsoft
3.8B params · text-generation · 697 likes · 296.3k downloads
Phi-4-mini-instruct is a 3.8B parameter model. At Q4 quantization it requires 2GB of VRAM. It runs comfortably on GeForce RTX 4090 (342 tok/s), GeForce RTX 5090 (513 tok/s), M4 Max 128GB (125 tok/s).
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 342 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 513 tok/s | comfortable |
| M4 Max 128GB | 128GB | 125 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 62 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 62 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 627 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 1261 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 302 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 250 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 242 tok/s | comfortable |