cnmoro/gliclass-large-v3.0-onnx

Original model here

Code for converting to onnx and quantizing here

Dependencies:

pip install huggingface-hub onnx onnxruntime numpy tokenizers

Inference code:

from huggingface_hub import hf_hub_download
from tokenizers import Tokenizer
import onnxruntime as ort
import numpy as np

class GLiClassOnnxInference:
    def __init__(self, model_id: str, use_int8_quant: bool = False):
        self.onnx_runtime_session = ort.InferenceSession(
            hf_hub_download(repo_id=model_id, filename="model_i8.onnx" if use_int8_quant else "model.onnx")
        )
        self.tokenizer = Tokenizer.from_file(
            hf_hub_download(repo_id=model_id, filename="tokenizer.json")
        )

    def encode(self, text: str, max_length: int = 512, pad: bool = True):
        encoded = self.tokenizer.encode(text)
        ids = encoded.ids
        mask = encoded.attention_mask
        if pad and len(ids) < max_length:
            pad_len = max_length - len(ids)
            ids += [self.tokenizer.token_to_id("[PAD]")] * pad_len
            mask += [0] * pad_len
        ids = ids[:max_length]
        mask = mask[:max_length]
        return np.array([ids], dtype=np.int64), np.array([mask], dtype=np.int64)

    def onnx_predict(self, text: str, labels: list[str]):
        full_text = "".join([f"<<LABEL>>{l}" for l in labels]) + "<<SEP>>" + text
        ids, mask = self.encode(full_text, max_length=512)
        ort_inputs = {"input_ids": ids, "attention_mask": mask}

        logits = self.onnx_runtime_session.run(None, ort_inputs)[0]
        probs = 1 / (1 + np.exp(-logits[0]))

        return [{"label": label, "score": float(prob)} for label, prob in zip(labels, probs)]

inference_session = GLiClassOnnxInference(
    "cnmoro/gliclass-large-v3.0-onnx",
    use_int8_quant = False
)
results = inference_session.onnx_predict(
    text = "One day I will see the world!",
    labels = ["travel", "dreams", "sport", "science", "politics"]
)
for r in results:
    print(f"{r['label']} => {r['score']:.3f}")