AWS Certified Cloud Practitioner
Get started today
Ultimate access to all questions.
Deep dive into the quiz with AI chat providers.
We prepare a focused prompt with your quiz and certificate details so each AI can offer a more tailored, in-depth explanation.
To avoid long-tail or irrelevant words, the team wants the model to restrict the sampling to the top 40 most likely tokens. Which setting should they apply?
Real Exam
Community
RRitesh
A
Set top-k to 40
B
Reduce top-p to 0.2
Explanation:
Explanation:
Top-k sampling is a technique that restricts the sampling to only the top k most likely tokens at each step of text generation. When you set top-k to 40, the model will only consider the 40 most probable tokens for sampling, effectively filtering out long-tail or irrelevant words that have lower probabilities.
Top-p (nucleus sampling) works differently - it selects from the smallest set of tokens whose cumulative probability exceeds a threshold p (e.g., 0.2). This doesn't guarantee a specific number of tokens and could potentially include more or fewer than 40 tokens depending on the probability distribution.
Therefore, to specifically restrict sampling to the top 40 most likely tokens, the correct approach is to set top-k to 40.
Comments
Loading comments...