gliner>=0.2.23 breaks HuggingFace Inference Endpoints (transformers 5.x incompatible with HF toolkit)

🚨 GLiNER returns uniform low scores with transformers v5.0.0 (released today)

mmBERT-base encoder instead of mdeberta-v3-base

Loading ONNX model requires config file

`IndexError` in `model.inference` when input contains empty strings

`InferencePackingConfig` fails when input contains empty strings

Can't load onnx pretrained model without pytorch model file

`ImportError: cannot import name 'amp' from 'apex'` on recent NVIDIA PyTorch Images (25.11)

Loading the model from huggingface is broken.

Difference GLiNER /urchade and /gliner-community