ataleckij/gliner_multi-v2.1_onnx

Temp fix for https://github.com/urchade/GLiNER/issues/314

1a. load ONNX FP32 (CPU)

from gliner import GLiNER

model = GLiNER.from_pretrained(
    model_id='ataleckij/gliner_multi-v2.1_onnx'
    map_location='cpu',
    load_tokenizer=True,
    load_onnx_model=True,
    onnx_model_file='model.onnx'
)
model.eval()

1b. load ONNX FP16 (GPU)

from gliner import GLiNER

model = GLiNER.from_pretrained(
    model_id='ataleckij/gliner_multi-v2.1_onnx'
    map_location='gpu',
    load_tokenizer=True,
    load_onnx_model=True,
    onnx_model_file='model_fp16.onnx'
)
model = model.half()
model.eval()

2. Inference

with torch.no_grad():
    preds_batch: list[list[dict]] = model.inference(
        texts,
        entity_types, 
        threshold=threshold, 
        batch_size=batch_size,
        flat_ner=False
    )