Octopus-v2
by NexaAI
2.5B params · text-generation · 890 likes · 350 downloads
Octopus-v2 is a 2.5B parameter model. At Q4 quantization it requires 1GB of VRAM. It runs comfortably on GeForce RTX 4090 (523 tok/s), GeForce RTX 5090 (785 tok/s), M4 Max 128GB (191 tok/s).
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 523 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 785 tok/s | comfortable |
| M4 Max 128GB | 128GB | 191 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 95 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 95 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 959 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 1930 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 463 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 383 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 371 tok/s | comfortable |