Pareto clarification

[arekay-nv]

The detailed definition of pareto submissions needs to be described. This includes the following:

• What models are going to be used by each submitters
• What dataset is going to be used for each model/submitter
• How are points defined (concurrency/latency/throughput) and what points are needed for submissions

Models

Current candidates are:

• Deepseek
• gpt-oss
• llama2-70b

Dataset

Ideally we would like dataset not currently used by MLPerf to avoid comparability.

Pareto points