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:
-
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
-
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.
