DeepSeek R1 8B
by DeepSeek · deepseek-r1 family
8B
parameters
text-generation code-generation reasoning math
DeepSeek R1 8B is a Llama 3.1-based distill of the full DeepSeek R1 reasoning model. It brings strong chain-of-thought reasoning to an 8B parameter size, making it accessible on consumer GPUs with 8-12 GB VRAM. Compared to the Qwen-based 7B distill, this Llama-based variant often shows better English performance. A solid choice for users who want reasoning capabilities without the VRAM requirements of the 14B or larger variants.
Quick Start with Ollama
ollama run 8b-q4_K_M | Creator | DeepSeek |
| Parameters | 8B |
| Architecture | transformer-decoder |
| Context | 128K tokens |
| Released | Jan 20, 2025 |
| License | MIT |
| Ollama | deepseek-r1:8b |
Quantization Options
| Format | File Size | VRAM Required | Quality | Ollama Tag |
|---|---|---|---|---|
| Q4_K_M rec | 5.2 GB | 7.5 GB | | 8b-q4_K_M |
| Q8_0 | 8.9 GB | 11.5 GB | | 8b-q8_0 |
| F16 | 16.5 GB | 20 GB | | 8b-fp16 |
Compatible Hardware
Q4_K_M requires 7.5 GB VRAM
Benchmark Scores
70.0
mmlu