Qwen2.5-VL-7B-Instruct
by Qwen
8.3B params · image-text-to-text · 1.5k likes · 4.6M downloads
Qwen2.5-VL-7B-Instruct is a 8.3B parameter model. At Q4 quantization it requires 4GB of VRAM. It runs comfortably on GeForce RTX 4090 (158 tok/s), GeForce RTX 5090 (237 tok/s), M4 Max 128GB (57 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Hyperbolic | 33 tok/s |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 158 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 237 tok/s | comfortable |
| M4 Max 128GB | 128GB | 57 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 28 tok/s | tight |
| M4 Pro 24GB | 24GB | 28 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 290 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 583 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 139 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 115 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 112 tok/s | comfortable |