adding length penalty to reward

[karthik-nexusflow]

Hi Team, While using the PPO pipeline we observe at times spikes in response length and were curious if any techniques related to length penalty is available or explored