temperature
scroll ↓ to Resources
Note
- parameter of the softmax function, when the base formula is modified so that the exponent is divided by a value T
- with T ⇒ infinity it increases entropy and flattens the distribution
- with T⇒ 0: more confident predictions become even more confident, less confident become even less confident
- this property allows controlling for diversity, creativity and consistency of tokens and is widely adopted by LLM decoding strategies
Resources
Links to this File
table file.inlinks, file.outlinks from [[]] and !outgoing([[]]) AND -"Changelog"