The vulnerability of the ensemble_classifer() function in the NVIDIA Megatron-LM training library allows attackers to execute arbitrary code, gain privileges, and disclose sensitive information.

The vulnerability of the ensembleclassifer function in the library for training large language models with NVIDIA Megatron-LM exists due to improper code generation control. Exploiting this vulnerability allows attackers to execute arbitrary code, gain elevated privileges, and expose sensitive...