ONNX runtime benchmarks

Compare execution backends over time

Explore model and operator performance across optimized, reference, and lightweight ONNX runtimes. Lower execution times are better.

Qwen2 BF16 ONNX Runtime CPU is intentionally unavailable. Missing onnx-light BF16 results instead indicate a benchmark failure and fail the Qwen2 shard before publication.

GenAI compatibility workflow status

Benchmark processors

Logical cores and instruction sets available to each benchmark worker.

Processor Logical cores Instruction sets
commits

Speedup vs ONNX Runtime (higher is better)