tag:github.com,2008:https://github.com/SemplificaAI/gliner2-rs/releases Tags from gliner2-rs 2026-05-23T10:24:38Z tag:github.com,2008:Repository/1213887246/v0.5.1 2026-05-23T10:24:38Z v0.5.1 <p>fix(ort): disable default features to stop download-binaries + copy-d…</p> <p>…ylibs</p> <p>Without `default-features = false`, ort downloads the full ONNX Runtime <br />distribution (a 617 MB CUDA EP DLL on x64 alone) and copies it next to <br />every consumer's release binary via `copy-dylibs`. Bundlers that <br />auto-include DLLs adjacent to the binary — Tauri's MSI packager being <br />the immediate motivator — pick those up too, inflating downstream <br />installers by 100+ MB for zero runtime benefit: consumers using <br />`load-dynamic` load onnxruntime.dll out of their own vendored libs/ <br />directory and never touch the EP DLLs ort dropped.</p> <p>The named EP features (qnn / cuda / rocm / coreml / openvino / <br />directml / tensorrt / xnnpack) stay on so the runtime fallback chain <br />in lib.rs and lib_v2.rs still compiles; an EP whose provider DLL is <br />absent from the loaded onnxruntime.dll fails registration and the <br />session quietly falls back to CPU — same observable behaviour as <br />before, just without the unused EP DLLs riding along to disk.</p> <p>Version bumped to 0.5.1.</p> dariofinardi tag:github.com,2008:Repository/1213887246/v0.5.0 2026-04-23T13:26:57Z v0.5.0 dariofinardi tag:github.com,2008:Repository/1213887246/v0.4.2 2026-04-23T05:57:54Z v0.4.2 <p>docs: integrate ARM64 FP32 Snapdragon benchmark data and switch prima…</p> <p>…ry index to Total Inference Time per Sentence across all tables</p> dariofinardi tag:github.com,2008:Repository/1213887246/v0.4.1 2026-04-22T20:51:13Z v0.4.1 <p>perf: optimize Hugging Face downloads to fetch only relevant V2 model…</p> <p>… variants based on OS</p> dariofinardi tag:github.com,2008:Repository/1213887246/v0.3.2 2026-04-21T15:12:12Z v0.3.2 dariofinardi tag:github.com,2008:Repository/1213887246/v0.3.1 2026-04-21T07:51:03Z v0.3.1 <p>fix(extraction): use character offsets for exact entity text slicing …</p> <p>…(v0.3.1)</p> dariofinardi tag:github.com,2008:Repository/1213887246/v0.3.0 2026-04-21T06:58:04Z v0.3.0 <p>docs(release): Release v0.3.0 with HF Hub download and dynamic CountL…</p> <p>…STM fix</p> dariofinardi