The objective is to make the generated summaries concise-that is, to control their length.
In the configuration of a generative AI model, particularly a large language model (LLM), the parameter used to directly control the maximum size of the response is the Output Length parameter (often referred to as max_output_tokens or max_tokens). By setting a low limit on this parameter, the team can ensure that the model is forced to terminate its response once that limit is reached, resulting in a shorter, more concise summary that is " easy to digest, " as requested.
The other parameters control different aspects of the output quality:
Temperature (C) controls the creativity or randomness of the output. Lowering it makes the output more predictable; raising it makes it more diverse. It does not control length.
Top-p (A) is a decoding method related to temperature that also controls the model ' s creativity by limiting the vocabulary from which it can choose the next token. It does not control length.
Safety settings (B) are used to filter and block the generation of harmful, illegal, or inappropriate content.
They do not affect the length or conciseness of the output.
(Reference: Google Cloud ' s Generative AI documentation on model parameters explicitly lists max_output_tokens or Output Length as the setting used to determine the maximum size of a model ' s generated response.)