Update display metric for E2E-QNA

[pgmpablo157321]

No description provided.

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[hans-intel]

Queries per sec (qps) is commonly used in single LLM call benchmarks, creating confusion why E2E-RAG has such a low perf number, especially we are using MLPerf GPT-OSS-120B model. In E2E-RAG,

• a single query can invoke more than a dozen LLM calls
• The pipeline involves 4 different inference models
• The pipeline involves non-model workload

Therefore, we propose to use tasks/sec, where each query represents a QnA task.

[hanyunfan]

LGTM