Local LLMs

**Running DeepSeek R1 on LLM Studio – Quick Start Guide**

> *DeepSeek R1* is a 13‑B parameter model that delivers strong performance on coding, reasoning and general LLM tasks. Below is a minimal setup to get it up and running in **LLM Studio** (the open‑source LLM dev hub).

| Step | Action | Command / Notes |
|——|——–|—————–|
| 1 | **Install LLM Studio** (if not already) | “`bashnpip install llm-studion“` |
| 2 | **Download the model** | “`bashnllm download DeepSeek/deepseek-r1n“` |
| 3 | **Load the model** | “`bashnllm run DeepSeek/deepseek-r1 –port 8000n“` |
| 4 | **Verify** – Open your browser to `http://localhost:8000` and hit the “Chat” tab. |
| 5 | **Optional GPU tuning** – If you have a CUDA‑capable GPU, set `–device cuda` to accelerate inference. |
| 6 | **Fine‑tune or prompt** – Use the built‑in Prompt Designer to experiment with different instruction styles. |

### Quick Tips

– **Memory**: DeepSeek R1 needs ~30 GB VRAM for full 2‑stage inference; use `–max_seq_len` to reduce context size if you’re on a smaller GPU.
– **Speed**: Enable `–batch_size 8` for better throughput during multi‑prompt sessions.
– **Safety**: Activate the built‑in content filter by adding `–safety` to the run command.

That’s it! You now have a fully functional DeepSeek R1 running locally in LLM Studio, ready for research or production prototyping. Happy modeling!