ResearchBenchmarksCode Generation on HumanEval (Throughput, Accuracy)Code Generation on HumanEval (Throughput, Accuracy)Follow1.91ThroughputHiSpec1.29641.45571.6151.7743Oct 1, 2025Evaluation resultsMethodMethodLinksThroughputAccuracyHiSpecBase Model=Llama3-8B,...Base Model=Llama3-8B, Accuracy Judge=Llama3.1-70B, Draft Model=None (single model)2025.101.91100SPRINTERBase Model=Llama3-8B,...Base Model=Llama3-8B, Accuracy Judge=Llama3.1-70B, Draft Model=Llama3.2-1B2025.101.3242