MLPerf Inference v5.0 open results — NVIDIA Jetson AGX Thor llama2-70b-99 Offline (raw log)

Fields

id

sources/mlperf-v5-0-nvidia-thor-llama2-70b-offline

URL

https://github.com/mlcommons/inference_results_v5.0/blob/main/open/NVIDIA/results/Thor_TRT/llama2-70b-99/Offline/performance/run_1/mlperf_log_summary.txt

Publisher

MLCommons / NVIDIA

Kind

benchmark

Published

2025-04-02

Accessed

2026-10-17

notes

Open-division submission (not closed). 1x NVIDIA Jetson AGX Thor Developer Kit 128GB, TensorRT 10.8, CUDA 12.8, FP16 weights. llama2-70b-99 and llama2-70b-99.9 share the same Offline performance log (Tokens per second 573.765, VALID). Quoted verbatim from mlperf_log_summary.txt on 2026-10-17. The 99 and 99.9 accuracy targets are two submissions of one run, so only one benchmark record is filed.

Notes

Open-division submission (not closed). 1x NVIDIA Jetson AGX Thor Developer Kit 128GB, TensorRT 10.8, CUDA 12.8, FP16 weights. llama2-70b-99 and llama2-70b-99.9 share the same Offline performance log (Tokens per second 573.765, VALID). Quoted verbatim from mlperf_log_summary.txt on 2026-10-17. The 99 and 99.9 accuracy targets are two submissions of one run, so only one benchmark record is filed.

Cross-references

This record references no other records.