[MLPerf Llama2] update llama2 reference fp32 accuracy

[nvzhihanj]

Address #1558

Old accuracy metrics: {'rouge1': 43.88, 'rouge2': 21.7108, 'rougeL': 28.2502, 'tokens_per_sample': 293.3}

New accuracy metrics: {'rouge1': 44.4312, 'rouge2': 22.0352, 'rougeL': 28.6162, 'tokens_per_sample': 294.45}

[github-actions]

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

[arjunsuresh]

Thank you for the update.

[pgmpablo157321]

@nvzhihanj Could you open a PR to modify the values here as well?

[FanhaiLu1]

@nvzhihanj Can you add a Accuracy Target for calibration dataset? I do prefer to get a quick accuracy result with a trustable small dataset.

[nvzhihanj]

@nvzhihanj Could you open a PR to modify the values here as well?

Opened mlcommons/inference_policies#289

[nvzhihanj]

@nvzhihanj Can you add a Accuracy Target for calibration dataset? I do prefer to get a quick accuracy result with a trustable small dataset.

@FanhaiLu1 Please get the reference log here: https://drive.google.com/file/d/14uTgBCGjMRKddSNAYkeFeKPtpl2-ft9m/view?usp=sharing. You can run a subset of it and compare to that subset with the reference log.