Reproducing AMD MLPerf Inference v6.1 Submission Results

AMD publishes a step-by-step guide to reproduce its MLPerf Inference v6.1 results on MI355X and MI350P GPUs. The post provides self-contained Docker images, quantized model weights, and specific benchmark recipes for workloads like Llama 2 70B and DeepSeek R1. This allows developers to independently verify performance claims using the provided hardware-specific tuning and accuracy constraints.