n8n Google PaLM Language Model nodeConfigure Google PaLM Language Model in n8n.

This is where the node learns what to call. The value names the model which will generate the completion, and it applies to every item the node processes in that execution.

• Model: the model which will generate the completion. Leave the stored value alone until a workflow gives you a reason to change it, or pass {{ $json.model }} when the choice travels with the incoming item.

Length gets its ceiling here. The field sets the maximum number of tokens to generate in the completion, tokens being the chunks a model reads and writes rather than whole words.

• Maximum Number of Tokens: the cap on generated text. Keep it tight for a one-line label, raise it for a paragraph, or read it from the item with {{ $json.maxTokens }}.

Predictability lives on this dial. It controls randomness: lowering it results in less random completions, and as the temperature approaches zero the model becomes deterministic and repetitive.

• Sampling Temperature: near zero for extraction and reformatting, where two runs on the same row should agree; higher when repetition is the thing you are trying to avoid.

Think of it as a filter on the candidate words. It is used to remove long tail low probability responses, the odd choices that make an otherwise usable answer wander.

• Top K: it defaults to -1, which disables it, so nothing is filtered until you set a value. Worth reaching for when completions drift off format rather than off topic.

Diversity gets its own control, through nucleus sampling. A value of 0.5 means half of all likelihood-weighted options are considered, so the pool the model draws from shrinks or widens with the number.

• Top P: the recommendation is to alter this or Sampling Temperature, but not both, since the two overlap and moving them together makes results hard to read.