CLI: How to Set the "--threads" Option

Certain processes within the Cognitive Toolkit have an option for --threads . To set this option to optimize performance of Cognitive Toolkit operations, it’s important to understand how and why it exists. 

--threads

The --threads option specifies the number of parallel processes to start.

Some important things to note:

• If a --threads option is NOT specified, the default is --threads 1.

• The --threads option is not forced. If set to --threads 1000, the operation applies as many threads as possible, “up to” the number specified.

Threads have different implications depending on the Cognitive Toolkit operation being run:

Setting the --threads value

If the chosen operation can be performed with 1 thread quickly, utilize the default setting.

If performance isn’t optimal at the default setting, try increasing the --threads value by intervals of two until the CPU or RAM is at capacity OR until diminishing returns are observed.

Examples

For CPU-intensive operations, the number of threads should be set to a lower number to improve performance.

addhashandextractedtext is a CPU-intensive operation. You may want to set the option to --threads 3

For lightweight operations, the number of threads can be set to a higher number to improve performance.

addapathvalidation is a lightweight operation. You may want to set the option to --threads 20.