Answer: temperature = 0.2, top-p = 0.7

## Explanation For legal contract summarization requiring **precise, repeatable language**, the optimal settings are: **Temperature = 0.2** (low temperature) **Top-p = 0.7** (moderate nucleus sampling) ### Why this combination works best: 1. **Low Temperature (0.2)**: - Temperature controls the randomness/creativity of the model's output - Lower values (closer to 0) make the model more deterministic and focused on the most likely tokens - For legal documents, we want minimal creativity and maximum consistency - Temperature of 0.2 ensures the model sticks closely to the most probable word choices 2. **Moderate Top-p (0.7)**: - Top-p (nucleus sampling) controls the diversity of token selection - Lower top-p values restrict the model to a smaller set of high-probability tokens - 0.7 allows some flexibility while still maintaining focus on the most relevant vocabulary - This prevents the model from considering too many unlikely alternatives ### Why other options are incorrect: - **Option B (temperature=0.9, top-p=1.0)**: Too high temperature introduces randomness, and top-p=1.0 considers all possible tokens, leading to inconsistent results - **Option C (temperature=1.2, top-p=1.0)**: Even higher temperature makes outputs highly creative and unpredictable, unsuitable for legal precision - **Option D (temperature=0.8, top-p=0.3)**: While top-p=0.3 is restrictive, temperature=0.8 is still too high for precise legal work **Key takeaway**: For tasks requiring precision and repeatability (like legal document processing), use low temperature values combined with moderate top-p settings to balance consistency with appropriate vocabulary selection.

Author: Ritesh Yadav