falcon-7b-instruct
by tiiuae
7.2B params · text-generation · 1.0k likes · 49.3k downloads
falcon-7b-instruct is a 7.2B parameter model. At Q4 quantization it requires 4GB of VRAM. It runs comfortably on GeForce RTX 4090 (181 tok/s), GeForce RTX 5090 (272 tok/s), M4 Max 128GB (66 tok/s).
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 181 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 272 tok/s | comfortable |
| M4 Max 128GB | 128GB | 66 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 33 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 33 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 333 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 670 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 160 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 133 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 129 tok/s | comfortable |