Temperature is the degree of randomness of model.

• If temperature=0, it will always choose the same highest frequency word as next generated.
• At higher temperature, it will choose less likely words, or less frequent word as next generated.
  When building models, prefer using temperature=0 for predictability & debugging
  When releasing to production, we might want to use higher temperature as higher variety of output