Sources
Loading...
Additional media
Loading...
DeepNewz, mobile.
People-sourced. AI-powered. Unbiased News.
Gemini has introduced several new API parameters, including logprobs, candidateCount, seed, presencePenalty, frequencyPenalty, and model version in response. These updates are expected to enhance the functionality and analytical capabilities of the Gemini API. The inclusion of logprobs will allow users to better understand the likelihood of tokens in a generated response, providing deeper insights into the model's decision-making process. More updates, including a slew of doc updates, are expected soon.
1. logprobs: The likelihood of tokens in a generated response. If "cat" has a logprob of -0.5 and "dog" has a logprob of -1.2, the model is more likely to choose "cat." Analyzing logprobs can help you to understand why the model picks one word over another, useful in tasks… https://t.co/wYFRODx1iZ
logprobs ftw https://t.co/kuVngWyyrv
logprobs you say ... https://t.co/oRaFKPPYgK https://t.co/UOzwwHGi3N
