Learning to Summarize: Best-of-N Selection
Beginner
The problem, from first principles
Best-of-N sampling draws several candidate summaries from the model, scores them with the reward model, and returns the best. It is a simple way to use a reward model at inference time and a baseline that the paper compares against.
From theory to code
Implement best_of_n_index(scores), returning the index of the candidate with the highest reward.
Constraints
- Return a Python
int.
Hints
Hint 1
Use np.argmax on the scores.
Keywords
research-papersreinforcement-learningalignmentrlhfsummarization
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.