2.8.0
Added:
- HowTo-Guide is created on RIB communication protocol usage explaining the new connection type, related cycle time parameter, necessary validations and preconfiguration.
Changed:
- When targeting AI Inference Server 2.8.0 or above, you no longer need to flatten model input tensors in pre-processing. The runtime now forwards and validates dimensional metadata automatically during GPU inference. Our E2E Object Detection Tutorial has been updated accordingly.
Security:
- onnx==1.20.0 - onnx ≤ 1.18.0 - Remote Path Traversal Vulnerability - CVE-2025-51480
- numpy==2.2.6
- pip==25.3 - pip ≤ 25.2 - Remote Symlink Traversal Vulnerability - CVE-2025-8869
- urllib3==2.6.3 - urllib3 1.0.x < 2.6.0 - Multiple Remote Denial of Service Vulnerabilities - GHSA-gm62-xv2j-4w53, GHSA-2xpw-w6gg-jr37
- fonttools==4.61.1
- keras==3.13.0
- werkzeug==3.1.5
- filelock==3.20.3
- torch==2.9.1+cpu
- torchvision==0.24.1+cpu
- protobuf==6.32.0
- onnxscript==0.5.7
- ml_dtypes==0.5.4
- onnx-ir==0.1.14
- setuptools==80.9.0