Lightweight multilingual grammar correction model, exported to ONNX and quantized to int8 for efficient inference.
It corrects text in multiple languages. Given a prompt like:
1import onnxruntime as ort
2import numpy as np
3
4# Load model
5session = ort.InferenceSession("hrm_grammar_light_int8.onnx")
6
7# Prepare your input sequence (tokenized, see below)
8input_ids = np.array([[...]], dtype=np.int64) # shape (1, seq_len)
9attention_mask = np.ones_like(input_ids)
10
11# Run inference
12outputs = session.run(["logits"], {
13 "input_ids": input_ids,
14 "labels": None,
15 "attention_mask": attention_mask,
16 "language_ids": None
17})
18logits = outputs[0] # (1, seq_len, vocab_size)
1from transformers import T5Tokenizer
2
3tokenizer = T5Tokenizer.from_pretrained("t5-small")
4prompt = "corregir español: el casa es grande"
5input_ids = tokenizer(prompt, return_tensors="np", padding="max_length", max_length=256, truncation=True)["input_ids"]
