Skip to main content

read_max_tokens

Function read_max_tokens 

Source
fn read_max_tokens(
    obj: &Map<String, Value>,
    issues: &mut Vec<FieldIssue>,
) -> Option<u32>
Expand description

Read max_tokens, which is u32 internally and -1 on the wire.

-1 is llama.cpp’s own idiom for “no limit”, and omitting the key means exactly that here — see with_hardcoded_defaults, which deliberately leaves max_tokens unset. So -1 is not an error to be reported, it is a spelling of a value this type already has. Any other negative is a client bug.