GreedySearch

Set GreedySearch to False to use an alternative algorithm for generating text.

By default, the generative model uses a greedy search algorithm to determine the output tokens. In this approach, the model chooses the token with the highest probability as the next output token.

You can set GreedySearch to False to use a multinomial sampling algorithm instead, which chooses the next token randomly based on a probability distribution over the entire vocabulary of the model. This option can sometimes give better results for longer sequences. However, the results vary for a particular input because the algorithm contains an element of randomness.

This parameter has an effect only when Type is set to GenerativeLLM.

Type: Boolean
Default: True
Required: No
Configuration Section: MyLLMModule
Example: GreedySerach=False
See Also: Type