- Change the name from
Requirements for models benchmarking[¶](http://localhost:8000/requirements/benchmark/#requirements-for-models-benchmarking)
- In general can drop the terms "Nexus package owner" / "package owner" throughout out. Often we can just us "user" or "experiment owner" or something more generic.
Terminology
Benchmark Target The AI model or algorithm being evaluated. This is the element that varies across benchmark experiments while the benchmark definition and benchmark instance specification are held constant.
This second part of this definition is incorrect (now a benchmark target may be == benchmark experiment). Also it uses a term "benchmark definition" that's not in the terminilolgy. Probably should say - for a given benchmark instance this is the primary element that is varied.
Benchmark Registration and Discoverability
REQ 2.3 uses the bare term benchmark and it's not clear what it is. From the context it sounds like it is a benchmark submission
REQ 2.3: Benchmark Registration The system must provide a method to define and register a benchmark:
The Benchmark Registration and Discoverability requirements miss the requirement adding logical benchmark. instance and the submission These are all someone bundled in current 2.3
Using the Benchmarking System
REQ 3.1 uses bare term benchmark.
Requirements for models benchmarking[¶](http://localhost:8000/requirements/benchmark/#requirements-for-models-benchmarking)Terminology
This second part of this definition is incorrect (now a benchmark target may be == benchmark experiment). Also it uses a term "benchmark definition" that's not in the terminilolgy. Probably should say - for a given benchmark instance this is the primary element that is varied.
Benchmark Registration and Discoverability
REQ 2.3 uses the bare term benchmark and it's not clear what it is. From the context it sounds like it is a benchmark submission
The
Benchmark Registration and Discoverabilityrequirements miss the requirement adding logical benchmark. instance and the submission These are all someone bundled in current 2.3Using the Benchmarking System
REQ 3.1 uses bare term benchmark.