DeepSeek-Coder-V2-Lite-Instruct
by deepseek-ai
15.7B params · text-generation · 557 likes · 205.0k downloads
DeepSeek-Coder-V2-Lite-Instruct is a 15.7B parameter model. At Q4 quantization it requires 8GB of VRAM. It runs comfortably on GeForce RTX 4090 (83 tok/s), GeForce RTX 5090 (125 tok/s), M4 Max 128GB (30 tok/s).
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 83 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 125 tok/s | comfortable |
| M4 Max 128GB | 128GB | 30 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 15 tok/s | tight |
| M4 Pro 24GB | 24GB | 15 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 153 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 308 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 73 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 61 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 59 tok/s | comfortable |