README.md

metadata

language:
  - en
  - de
  - es
  - fr
  - it
  - nl
  - pt
license: apache-2.0
tags:
  - sentence-classification
  - text-classification
  - onnx
  - multilingual
datasets:
  - TigreGotico/sentence-types-multilingual

sentence-types

Multilingual sentence-type classifiers (ONNX) trained on TigreGotico/sentence-types-multilingual (9,900 balanced samples per language, 6 classes).

Used by little_questions.

Classes

command, exclamation, polar_question, request, statement, wh_question

Models

File	Language
sentence_type_EN_0.8.0.onnx	English
sentence_type_DE_0.8.0.onnx	German
sentence_type_ES_0.8.0.onnx	Spanish
sentence_type_FR_0.8.0.onnx	French
sentence_type_IT_0.8.0.onnx	Italian
sentence_type_NL_0.8.0.onnx	Dutch
sentence_type_PT_0.8.0.onnx	Portuguese

Accuracy

Language	Accuracy	Macro F1
EN	99.2%	99.2%
NL	98.8%	98.8%
FR	97.1%	97.1%
IT	97.0%	97.0%
PT	95.4%	95.4%
DE	85.6%	84.9%
ES	74.6%	72.7%

Inference

import onnxruntime as rt, numpy as np, json

sess = rt.InferenceSession("sentence_type_EN_0.8.0.onnx")
classes = json.loads(sess.get_modelmeta().custom_metadata_map["classes"])
inp = np.array(["Who invented the telephone?"], dtype=object)
label_idx, probs = sess.run(None, {"input": inp})
print(classes[int(label_idx[0])])   # wh_question