ResearchTasksLLM Training PerformanceLLM Training PerformanceFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedAlpaca Qwen3 32Btorchtune + AC + LCE38.43Memory (GB/GPU)6May 21, 2026Qwen models on RTX 3090 (24GB VRAM) 2.5 (train)MegaTrain35.09TFLOPS5Apr 8, 2026Qwen models on A6000 (48GB VRAM) 2.5 (train)MegaTrain56.82TFLOPS5Apr 8, 2026Alpaca Llama 3.3 70Btorchtune + AC74.75Memory Usage (GB/GPU)3May 21, 2026