Qwen2-1.5B-Instruct
by Qwen
1.5B params · text-generation · 161 likes · 3.2M downloads
Qwen2-1.5B-Instruct is a 1.5B parameter model. At Q4 quantization it requires 1GB of VRAM. It runs comfortably on GeForce RTX 4090 (850 tok/s), GeForce RTX 5090 (1275 tok/s), M4 Max 128GB (311 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Featherless |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 850 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 1275 tok/s | comfortable |
| M4 Max 128GB | 128GB | 311 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 155 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 155 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 1558 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 3134 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 752 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 621 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 603 tok/s | comfortable |