DeBERTa commited on
Commit
a1052d3
1 Parent(s): 1d31a10

Update README.md

Browse files
Files changed (2) hide show
  1. README.md +7 -8
  2. pytorch_model.bin +1 -1
README.md CHANGED
@@ -8,17 +8,16 @@ thumbnail: https://huggingface.co/front/thumbnails/microsoft.png
8
  license: mit
9
  ---
10
 
11
- ## DeBERTa: Decoding-enhanced BERT with Disentangled Attention
12
 
13
  [DeBERTa](https://arxiv.org/abs/2006.03654) improves the BERT and RoBERTa models using disentangled attention and enhanced mask decoder. With those two improvements, DeBERTa out perform RoBERTa on a majority of NLU tasks with 80GB training data.
14
 
15
- Please check the [official repository](https://github.com/microsoft/DeBERTa) for more details and updates.
16
 
17
- In DeBERTa V3, we replaced the MLM objective with the RTD(Replaced Token Detection) objective introduced by ELECTRA for pre-training, as well as some innovations to be introduced in our upcoming paper. Compared to DeBERTa-V2, our V3 version significantly improves the model performance in downstream tasks. You can find a simple introduction about the model from the appendix A11 in our original [paper](https://arxiv.org/abs/2006.03654), but we will provide more details in a separate write-up.
18
 
19
- mDeBERTa is the multilingual version of DeBERTa with the same model structure but was trained on the CC100 multilingual data.
20
-
21
- The mDeBERTa V3 base model comes with 12 layers and a hidden size of 768. Its total parameter number is 280M since we use a vocabulary containing 250K tokens which introduce 190M parameters in the Embedding layer. This model was trained using the 2.5T CC100 data as XLM-R.
22
 
23
 
24
  #### Fine-tuning on NLU tasks
@@ -68,7 +67,7 @@ python -m torch.distributed.launch --nproc_per_node=${num_gpus} \
68
 
69
  ### Citation
70
 
71
- If you find DeBERTa useful for your work, please cite the following paper:
72
 
73
  ``` latex
74
  @misc{he2021debertav3,
@@ -90,4 +89,4 @@ booktitle={International Conference on Learning Representations},
90
  year={2021},
91
  url={https://openreview.net/forum?id=XPZIaotutsD}
92
  }
93
- ```
 
8
  license: mit
9
  ---
10
 
11
+ ## DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing
12
 
13
  [DeBERTa](https://arxiv.org/abs/2006.03654) improves the BERT and RoBERTa models using disentangled attention and enhanced mask decoder. With those two improvements, DeBERTa out perform RoBERTa on a majority of NLU tasks with 80GB training data.
14
 
15
+ In [DeBERTa V3](https://arxiv.org/abs/2111.09543), we further improved the efficiency of DeBERTa using ELECTRA-Style pre-training with Gradient Disentangled Embedding Sharing. Compared to DeBERTa, our V3 version significantly improves the model performance on downstream tasks. You can find more technique details about the new model from our [paper](https://arxiv.org/abs/2111.09543).
16
 
17
+ Please check the [official repository](https://github.com/microsoft/DeBERTa) for more implementation details and updates.
18
 
19
+ mDeBERTa is multilingual version of DeBERTa which use the same structure as DeBERTa and was trained with CC100 multilingual data.
20
+ The mDeBERTa V3 base model comes with 12 layers and a hidden size of 768. It has 86M backbone parameters with a vocabulary containing 250K tokens which introduces 190M parameters in the Embedding layer. This model was trained using the 2.5T CC100 data as XLM-R.
 
21
 
22
 
23
  #### Fine-tuning on NLU tasks
 
67
 
68
  ### Citation
69
 
70
+ If you find DeBERTa useful for your work, please cite the following papers:
71
 
72
  ``` latex
73
  @misc{he2021debertav3,
 
89
  year={2021},
90
  url={https://openreview.net/forum?id=XPZIaotutsD}
91
  }
92
+ ```
pytorch_model.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:05c748186a22f523505099ce137f90dd4e55f875a4035c11350aaa125932230c
3
  size 560166373
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:92671f84f9fd084a1365279f3c607bea6a7aef500ef3fcb28e03ac35c651fc30
3
  size 560166373