Releases: justinchuby/onnx-genai
Release list
v0.1.0-dev.6
Development release tagged from exact candidate 49590f7c22da2ee554aa131e394e8c5f50ef7961.
Publishes 40 Rust crates at 0.1.0-dev.6: the reviewed 36-crate runtime set plus four package-time test dependencies required by Cargo (onnx-runtime-einsum-conformance, onnx-runtime-hostmon, onnx-runtime-operator-selection, and onnx-genai-paged-attention).
PyPI 0.1.0.dev6 distributions: nxrt, onnx-genai, and onnx-genai-server. Linux x86_64, Windows AMD64, and macOS arm64 wheels are provided; GenAI packages additionally include Windows ARM64. onnx-std has no Python publish workflow in this release.
The separate nxrt-ep-v0.1.0-dev.6 prerelease publishes CPU/CUDA ORT plugin-EP packages. Mobius producer conformance remains manual-only.
nxrt-ep-v0.1.0-dev.6
Development release of the ORT plugin-EP packages, tagged from exact candidate 49590f7c22da2ee554aa131e394e8c5f50ef7961.
Publishes nxrt-ep-cpu==0.1.0.dev6 for Linux x86_64, Windows AMD64, Windows ARM64, and macOS arm64, and experimental nxrt-ep-cuda==0.1.0.dev6 for Linux x86_64. CUDA runtime libraries are dynamically loaded rather than bundled.
Release wheels are externally available on PyPI; corresponding build artifacts remain attached to Actions run 34449874469.
v0.1.0-dev.1
This pre-release/dev tag cuts the onnx-runtime-* runtime layer at 0.1.0-dev.1.
New since v0.1.0-dev.0: the onnx-runtime-tracer crate, a unified pure-Rust TraceCollector fan-out supporting Chrome Trace/JSONL/Perfetto export plus optional CUPTI GPU-kernel tracing and Intel ITT; and the full ORT2 runtime layer: ir, loader, session, optimizer, ep-api, ep-cpu, ep-cuda, eager, capi, cpuinfo, shape-inference, and python.
Purpose: lets external execution-provider crates, including the MLX EP, depend on onnx-runtime-tracer via this git tag.