Qwen2.5-7B-Instruct
by Qwen
7.6B params · text-generation · 1.1k likes · 22.3M downloads
Qwen2.5-7B-Instruct is a 7.6B parameter model. At Q4 quantization it requires 4GB of VRAM. It runs comfortably on GeForce RTX 4090 (172 tok/s), GeForce RTX 5090 (258 tok/s), M4 Max 128GB (63 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Together AI | 139 tok/s | ||
| Featherless |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 172 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 258 tok/s | comfortable |
| M4 Max 128GB | 128GB | 63 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 31 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 31 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 315 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 635 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 152 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 126 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 122 tok/s | comfortable |