Views
No views yet
1import torch
2from transformers import RoFormerForMaskedLM, RoFormerTokenizer
3
4text = "今天[MASK]很好，我[MASK]去公园玩。"
5tokenizer = RoFormerTokenizer.from_pretrained("junnyu/roformer_chinese_char_small")
6pt_model = RoFormerForMaskedLM.from_pretrained("junnyu/roformer_chinese_char_small")
7pt_inputs = tokenizer(text, return_tensors="pt")
8with torch.no_grad():
9 pt_outputs = pt_model(**pt_inputs).logits[0]
10pt_outputs_sentence = "pytorch: "
11for i, id in enumerate(tokenizer.encode(text)):
12 if id == tokenizer.mask_token_id:
13 tokens = tokenizer.convert_ids_to_tokens(pt_outputs[i].topk(k=5)[1])
14 pt_outputs_sentence += "[" + "||".join(tokens) + "]"
15 else:
16 pt_outputs_sentence += "".join(
17 tokenizer.convert_ids_to_tokens([id], skip_special_tokens=True))
18print(pt_outputs_sentence)
19# pytorch: 今天[也||都||又||还||我]很好，我[就||想||去||也||又]去公园玩。1import tensorflow as tf
2from transformers import RoFormerTokenizer, TFRoFormerForMaskedLM
3text = "今天[MASK]很好，我[MASK]去公园玩。"
4tokenizer = RoFormerTokenizer.from_pretrained("junnyu/roformer_chinese_char_small")
5tf_model = TFRoFormerForMaskedLM.from_pretrained("junnyu/roformer_chinese_char_small")
6tf_inputs = tokenizer(text, return_tensors="tf")
7tf_outputs = tf_model(**tf_inputs, training=False).logits[0]
8tf_outputs_sentence = "tf2.0: "
9for i, id in enumerate(tokenizer.encode(text)):
10 if id == tokenizer.mask_token_id:
11 tokens = tokenizer.convert_ids_to_tokens(
12 tf.math.top_k(tf_outputs[i], k=5)[1])
13 tf_outputs_sentence += "[" + "||".join(tokens) + "]"
14 else:
15 tf_outputs_sentence += "".join(
16 tokenizer.convert_ids_to_tokens([id], skip_special_tokens=True))
17print(tf_outputs_sentence)
18# tf2.0: 今天[也||都||又||还||我]很好，我[就||想||去||也||又]去公园玩。1@misc{su2021roformer,
2 title={RoFormer: Enhanced Transformer with Rotary Position Embedding},
3 author={Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu},
4 year={2021},
5 eprint={2104.09864},
6 archivePrefix={arXiv},
7 primaryClass={cs.CL}
8}