Examples

Tutorial¶

Our short tutorial, Taking a random walk, introduces core ado concepts and is the recommended place to start.

General Examples¶

The following examples illustrate general features of ado. They build on the concepts learned in the tutorial and leverage pre-existing data and/or toy measurements allowing them to run quickly.

• Search a space with an optimizer
• Search a space based on a custom objective function
• Identify the important dimensions of a space

After following these examples you can also try applying capabilities learned in one example to another.

Foundation Models Characterization¶

The following examples illustrate using the vllm_performance and SFTTrainer actuators which offer benchmarking experiments for foundation model inference and fine-tuning respectively.

• Measure throughput of fine-tuning locally
• Measure throughput of fine-tuning on a RayCluster with GPUs
• Find the request rate giving the highest stable throughput for an inference server

Adding experiments or analysis tools to ado¶

The search a space based on a custom objective function example, combines with the creating a custom experiment documentation to illustrate a simple method for adding your own experiments to ado.

For adding actuators, we provide an example template actuator repository which can be used with our documentation on writing actuators.

For adding operators, we have an example template operator repository which can be used with our documentation on writing operators.

What's next¶