Configure Kubeflow on Google Kubernetes Engine to handle training jobs via TFJob, ensuring compatibility with TensorFlow-based frameworks but potentially limiting support for others.

Utilize the AI Platform's custom containers feature to accept training jobs across any framework, including custom libraries, offering flexibility and reducing management overhead.

Establish a Slurm workload manager for scheduling and executing jobs on your cloud infrastructure, which requires significant setup and maintenance but offers fine-grained control.

Develop a library of VM images on Compute Engine, making them available in a centralized repository, which could increase costs and complexity due to the need for multiple images for different frameworks.

Implement a combination of AI Platform's custom containers for most frameworks and Kubeflow for TensorFlow-based jobs, ensuring broad compatibility and optimized performance where possible. Choose two correct options.