inferenceMaxTokens

Maximum number of tokens the model can generate in response to each prompt during RFT training.