Skip to content

Releases: lemonade-sdk/ort-server

0.3.7

Choose a tag to compare

@github-actions github-actions released this 15 Jul 16:23
Filter special tokens from RobertaProcessing / BertProcessing post-pr…

0.3.6

Choose a tag to compare

@github-actions github-actions released this 14 Jul 16:52
Complete the tokenizer contract: filter special tokens; require confi…

0.3.5

Choose a tag to compare

@github-actions github-actions released this 14 Jul 00:59
check_deps: allow stock macOS system frameworks

CoreFoundation et al. ship with the OS; the checker was flagging them.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

0.3.4

Choose a tag to compare

@github-actions github-actions released this 13 Jul 23:39
Fail cleanly on a corrupt tokenizer.json instead of panicking across …

0.3.3

Choose a tag to compare

@github-actions github-actions released this 13 Jul 22:46
Strip tokenizer padding before inference (served scores were wrong)

A tokenizer.json with a `padding` section — which real HuggingFace repos ship,
including our own published phishing classifiers — makes the Rust tokenizer pad
every encoding to a fixed width. transformers does not pad by default. We
forwarded those [PAD] ids to the model under an all-ones attention mask, so it
attended to ~500 padding tokens as if they were text.

Impact: every score served for the published catalog models was wrong. On the
probe 'Account notice: sign in to review recent activity.' the real model
returned LABEL_0=0.0005 where HuggingFace returns 0.7474. Affects 0.2.0-0.3.2
(0.3.2 was never released).

- drop trailing pad ids (pad_id read from tokenizer.json) before building the
  inputs, reproducing the HF reference exactly (now 0.7474327 vs 0.747434)
- new tiny-pad fixture: tiny-clf with padding enabled in tokenizer.json; its
  golden scores are tiny-clf's, so any regression that lets pad ids through
  fails immediately. The existing fixtures could not catch this — their
  tokenizers declare no padding, which is why it shipped.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

0.3.1

Choose a tag to compare

@github-actions github-actions released this 13 Jul 20:04
Self-review + round-2 fixes: special tokens, contract hardening, lice…

0.3.0

Choose a tag to compare

@github-actions github-actions released this 13 Jul 16:07
Stage: skip the macOS .dSYM directory when copying libonnxruntime

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

0.2.0

Choose a tag to compare

@ramkrishna2910 ramkrishna2910 released this 10 Jul 21:33

ort-server 0.2.0 (ONNX Runtime 1.27.0). Retag of v0.2.0 to match the backend version pin (tag == version, no v prefix).

v0.1.1

Choose a tag to compare

@github-actions github-actions released this 08 Jul 22:39
Define ort-extensions op set by scope: full text ops, exclude vision/…

v0.1.0

Choose a tag to compare

@github-actions github-actions released this 08 Jul 20:04
Use runner-default CMake + policy flag via bash (fix Windows VS 2026)

windows-latest migrated to Visual Studio 2026, which needs the runner's
CMake 4.x; pinning 3.31 broke generator detection. Drop the pin, let CMake
auto-detect VS 2026, and pass CMAKE_POLICY_VERSION_MINIMUM=3.5 through bash
(PowerShell had mangled it to '3') so old subprojects still configure.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>