Models / deepseek-coder
deepseek-coder by deepseek-ai
6 models (1.3B to ~236B) · ↓ 702.6K downloads a month · from $0.10/h on RTX 3060 right now. Each model's versions do the same job; quantized ones (AWQ, FP8, 4-bit) need less GPU memory and run on cheaper machines.
deepseek-coder 1.3B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| Originaldeepseek-ai/deepseek-coder-1.3b-instruct | vLLM | 2.7 GB | 7 GB | $0.10/h RTX 3060 |
deepseek-coder 6.7B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| Originaldeepseek-ai/deepseek-coder-6.7b-base | vLLM | 13 GB | 22 GB | $0.24/h RTX 3090 |
deepseek-coder 6.7B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| Originaldeepseek-ai/deepseek-coder-6.7b-instruct | vLLM | 13 GB | 22 GB | $0.24/h RTX 3090 |
deepseek-coder 7B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| Originaldeepseek-ai/deepseek-coder-7b-instruct-v1.5 | vLLM | 14 GB | 19.8 GB | $0.18/h 2x RTX 3060 |
deepseek-coder 33B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| Originaldeepseek-ai/deepseek-coder-33b-instruct | vLLM | 67 GB | 78.6 GB | $0.89/h 2x Nvidia A40 |
| GPTQ (TheBloke)TheBloke/deepseek-coder-33B-instruct-GPTQ | vLLM | 17 GB | 23.9 GB | $0.24/h RTX 3090 |
DeepSeek-Coder ~236B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| Instructdeepseek-ai/DeepSeek-Coder-V2-Instruct | vLLM | 471 GB | 537.3 GB | $11.77/h 8x RTX PRO 6000 |
| Instruct 0724deepseek-ai/DeepSeek-Coder-V2-Instruct-0724 | vLLM | 471 GB | 537.3 GB | $11.77/h 8x RTX PRO 6000 |
Model details come from Hugging Face and are refreshed daily; prices follow the live offers. GPU memory is GridPup's estimate: the weights, room for an 8,192-token conversation (chat models) and the runtime's own memory. Large models are split across the GPUs of one machine ("4x RTX 4090").