pypi / gliner

gliner

What changed in gliner 0.2, from its official release notes, upgrade guides and advisories. Sources checked 2026-10-24.

3 breaking16 new APIs and options2 performance13 fixes and other

See which of these apply to your code, ranked, in your agent:

npx @unstale/cli init

gliner 0.234 changes

Install the separate onnx package when exporting to ONNX.
0.2.29release notes
Fails fast when class_token_index and vocab_size are inconsistent.
0.2.28release notes
Loads from_pretrained at the target dtype.
0.2.27release notes

Install this extra for GPU ONNX Runtime.
0.2.29release notes
Install this extra for CPU ONNX Runtime.
0.2.29release notes
Install this extra for OpenVINO.
0.2.29release notes
Select OpenVINO or ONNX Runtime when loading a model while retaining the familiar prediction API.
0.2.29release notes
Pass label-to-description dictionaries to inference or serving; descriptions become prompts while predictions retain label names, and inputs can have separate dictionaries.
0.2.29release notes
Export supported PyTorch models directly to OpenVINO IR for deployment.
0.2.29release notes
Optionally weight positive spans by their width during training.
0.2.29release notes
Request matched-label representations for downstream processing; vector outputs require native PyTorch inference.
0.2.29release notes
Request entity and relation representations for downstream processing; vector outputs require native PyTorch inference.
0.2.29release notes
Add span-level Dice loss as an additional training objective.
0.2.29release notes
Added PolyLoRA support.
0.2.28release notes
Added support for streaming inference.
0.2.28release notes
Adds prompt compression.
0.2.27release notes
Adds quantization support and restricts quantize= to int8.
0.2.27release notes
Adds serving support.
0.2.27release notes
Adds variant= to from_pretrained for selective half-precision downloads.
0.2.27release notes

Adds compile support for an approximately 1.9x GPU speedup.
0.2.27release notes
Adds low_cpu_mem_usage=True to from_pretrained for approximately 2x faster cold starts.
0.2.27release notes

Fix crashes and incorrect label assignments when batching texts with different label sets.
0.2.29release notes
Propagate local_files_only through model initialization and save or load auxiliary tokenizers locally for improved offline loading.
0.2.29release notes
Fix best-checkpoint restoration and resuming training from native GLiNER checkpoints.
0.2.29release notes
Fix training on negative examples that have no annotated entities.
0.2.29release notes
Forward relation types and handle entity representations correctly during evaluation.
0.2.29release notes
Preserve warmup_ratio compatibility with newer Transformers versions.
0.2.29release notes
Preserves CLS/SEP tokens for compatible tokenizers on Transformers v5.
0.2.28release notes
Fixed issues related to Docker serving.
0.2.28release notes
Fixed handling of the local_files_only flag.
0.2.28release notes
Fixed model training and saving.
0.2.28release notes
Fixed inference for multi-task models.
0.2.28release notes
Fixed token_lengths leaking into the labels encoder.
0.2.28release notes
Adds a proper GLiNER client and fixes bugs.
0.2.27release notes