| # Local DeepSeek (R1 family) for the modding agent. | |
| # Build: ollama create modding-deepseek -f Modelfile.deepseek | |
| # Run: python modding_agent.py --model modding-deepseek --text-tools --fold-system --temperature 0.6 \ | |
| # --allow "C:/Games" | |
| # | |
| # Pick a size that fits your VRAM: deepseek-r1:8b / 14b / 32b / 70b (671b needs a server). | |
| # Check capabilities with: ollama show deepseek-r1:14b (look for "tools" and "thinking") | |
| FROM deepseek-r1:14b | |
| # Your own GGUF instead? Bare GGUF files have no chat template, so copy the TEMPLATE and stop | |
| # parameters from the library model of the same family first: | |
| # ollama show --modelfile deepseek-r1:14b | |
| # paste them below and change FROM to: FROM ./your-deepseek.gguf | |
| PARAMETER num_ctx 8192 | |
| # DeepSeek's published advice for R1: temperature around 0.6, and no separate system prompt | |
| # (that is what --fold-system does: it moves the instructions into the first user message). | |
| PARAMETER temperature 0.6 | |
| PARAMETER top_p 0.95 | |