v0.0.6

What's Changed

• Upgrade the package baseline to Swift 6.3+.
• Update documented/runtime baseline for TensorRT 11.x and CUDA 13.3.
• Remove the legacy weak precision builder-flag path from the native ONNX builder.
• Treat Precision.fp16 as model precision intent instead of calling removed TensorRT flags; use typed ONNX graphs, ModelOpt AutoCast, or Q/DQ nodes for reduced precision.
• Move CI container metadata to the newest verified TensorRT NGC tag, 26.05-py3.
• Clean up local/macOS development imports and Linux build warnings.

Validation

• Local: swift build --target TensorRT and swift build --target TensorRTTests.
• Ubuntu RTX 4090 host: Swift 6.3.2, TensorRT 11.1/CUDA 13.3, swift test passed with 23 tests and no compiler warnings.

v0.0.4

• Add stateless host execution path for TensorRT < 10 (enqueueV2 bindings).
• Fallback to stateless execution when persistent contexts are unavailable.

v0.0.3

What's New in v0.0.3

CI/CD Improvements

• Added GitHub Actions CI workflow with self-hosted GPU runner (RTX 4090)
• Docker-based build environment with Swift 6.2.3 + TensorRT 10.7.0
• All 17 integration tests passing against real NVIDIA TensorRT APIs

Bug Fixes

• Fixed Swift compiler warnings in test files
• Fixed CI permission issues with Docker-created files

Infrastructure

• Added CI status badge to README
• Added version badge to README

0.0.2

What's Changed

• Reuse persistent TensorRT execution context by @zamderax in #1
• Dynamic ONNX builds with optimization profiles by @zamderax in #9

New Contributors

• @zamderax made their first contribution in #1

Full Changelog: 0.0.1...0.0.2

0.0.1

Initial WIP release for Linux + TensorRT system integration.