DeepSeek R1 Official System Prompt Guide

For DeepSeek-R1 (Reasoning Model):
- Do NOT use system prompts — place all instructions in the user message
- Do NOT use few-shot examples (they confuse the reasoning chain)
- Set temperature to 0.6 for optimal reasoning
- Let the model reason independently

For DeepSeek-V3 (Chat Model):
- Use system prompts to define behavior and persona
- Few-shot examples work well
- Temperature 0.7-1.0 for creative tasks, 0.3-0.5 for precise tasks

General Settings:
- max_tokens: Set appropriately for your task
- top_p: 0.95 recommended
- frequency_penalty: 0 (default)
- presence_penalty: 0 (default)