Orkhon Versions Save

This release comes with:

• ONNX interface
• New asynchronous servicing methods
• Shareable server runtime
• Nuclei asynchronous runtime
• Inferring input facts for frozen model
• Improves throughput:
  • ~4.8361 GiB/s prediction throughput
  • 3_000 concurrent requests take ~4ms on average