• Compatibility: This version requires EVA App v2.7.0 or higher.

Performance Enhancements

• IG Inference Runtime Optimization: Refactored IG embedding and inference paths into reusable core modules, and improved serving efficiency by applying torch.compile(dynamic=True) and CUDA bfloat16 execution with eager fallback.
• IG API validation and documentation enhancement: Added proxy-side validation for IG inference payload and documented get_embedding / inference request-response format including scenario threshold and interest zone.
• Adjusted default serving worker settings: Reduced the default number of workers for OmDet and MMGDino from 2 to 1 for a lighter default runtime profile.

Bug Fixes

• IG runtime packaging and utility loading fixes: Included new IG utility module in model archive flow and improved interest-zone utility import method to ensure IG runtime dependency is resolved correctly in TorchServe.
• IG box conversion compatibility fix: Replaced the use of torchvision.ops.box_convert in IG hot path with Torch-native conversion helper to prevent compile and runtime compatibility issues.