Dolphin-Mistral-24B-Venice-Edition
by dphn
24.0B params · text-generation · 580 likes · 9.0k downloads
Dolphin-Mistral-24B-Venice-Edition is a 24.0B parameter model. At Q4 quantization it requires 12GB of VRAM. It runs comfortably on GeForce RTX 4090 (54 tok/s), GeForce RTX 5090 (82 tok/s), A100 PCIe 80 GB (100 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Featherless |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 54 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 82 tok/s | comfortable |
| M4 Max 128GB | 128GB | 20 tok/s | tight |
| M4 Pro 48GB | 48GB | 10 tok/s | tight |
| M4 Pro 24GB | 24GB | 10 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 100 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 201 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 48 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 39 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 38 tok/s | comfortable |