ssakethch/h200-quantization-benchmarks

H200量化基准测试数据集包含40个量化和非量化的指令调优LLM在NVIDIA H200 MIG设置下的基准测试结果。该数据集支持在准确性和吞吐量维度上对量化方法（AWQ、GPTQ、fp8、bf16）进行可重复的比较。数据集包含多个配置，如准确性和吞吐量的测试结果、模型清单、以及领导板排名等。硬件配置为NVIDIA H200 SXM（141 GB HBM3e），使用MIG分区实例和vLLM推理引擎。评估任务包括ARC Challenge、HellaSwag、Winogrande、GSM8K数学推理和MMLU等。

Benchmark results for 40 quantized and non-quantized instruction-tuned LLMs evaluated on an NVIDIA H200 MIG (Multi-Instance GPU) setup. This dataset supports reproducible comparison of quantization methods (AWQ, GPTQ, fp8, bf16) across accuracy and throughput dimensions. The dataset includes multiple configurations such as accuracy and throughput test results, model inventory, and leaderboard rankings. Hardware configuration includes NVIDIA H200 SXM (141 GB HBM3e) with MIG partitioned instances and vLLM inference engine. Evaluation tasks include ARC Challenge, HellaSwag, Winogrande, GSM8K math reasoning, and MMLU.