Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,58 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
---
|
| 4 |
+
# Disclamer
|
| 5 |
+
I do not own, distribute, or take credits for this model, all copyrights belong to [Instadeep](https://huggingface.co/InstaDeepAI) under the [MIT licence](https://github.com/instadeepai/tunbert/)
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
# how to load the model
|
| 9 |
+
```python
|
| 10 |
+
!git clone https://huggingface.co/not-lain/TunBERT
|
| 11 |
+
!pip install transformers
|
| 12 |
+
import torch.nn as nn
|
| 13 |
+
import torch
|
| 14 |
+
from transformers import AutoTokenizer, AutoModelForMaskedLM, AutoModelForSequenceClassification, PreTrainedModel,AutoConfig, BertModel
|
| 15 |
+
from transformers.modeling_outputs import SequenceClassifierOutput
|
| 16 |
+
config = AutoConfig.from_pretrained("not-lain/TunBERT")
|
| 17 |
+
class classifier(nn.Module):
|
| 18 |
+
def __init__(self,config):
|
| 19 |
+
super().__init__()
|
| 20 |
+
|
| 21 |
+
self.layer0 = nn.Linear(in_features=config.hidden_size, out_features=config.hidden_size, bias=True)
|
| 22 |
+
self.layer1 = nn.Linear(in_features=config.hidden_size, out_features=config.type_vocab_size, bias=True)
|
| 23 |
+
def forward(self,tensor):
|
| 24 |
+
out1 = self.layer0(tensor)
|
| 25 |
+
return self.layer1(out1)
|
| 26 |
+
|
| 27 |
+
|
| 28 |
+
class TunBERT(PreTrainedModel):
|
| 29 |
+
def __init__(self, config):
|
| 30 |
+
super().__init__(config)
|
| 31 |
+
self.BertModel = BertModel(config)
|
| 32 |
+
self.dropout = nn.Dropout(p=0.1, inplace=False)
|
| 33 |
+
self.classifier = classifier(config)
|
| 34 |
+
|
| 35 |
+
def forward(self,input_ids=None,token_type_ids=None,attention_mask=None,labels=None) :
|
| 36 |
+
outputs = self.BertModel(input_ids,token_type_ids,attention_mask)
|
| 37 |
+
sequence_output = self.dropout(outputs.last_hidden_state)
|
| 38 |
+
logits = self.classifier(sequence_output)
|
| 39 |
+
loss =None
|
| 40 |
+
if labels is not None :
|
| 41 |
+
loss_func = nn.CrossentropyLoss()
|
| 42 |
+
loss = loss_func(logits.view(-1,self.config.type_vocab_size),labels.view(-1))
|
| 43 |
+
return SequenceClassifierOutput(loss = loss, logits= logits, hidden_states=outputs.last_hidden_state,attentions=outputs.attentions)
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
tunbert = TunBERT(config)
|
| 47 |
+
tunbert.load_state_dict(torch.load("/content/TunBERT/pytorch_model.bin"))
|
| 48 |
+
|
| 49 |
+
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
|
| 50 |
+
```
|
| 51 |
+
|
| 52 |
+
|
| 53 |
+
# how to use the model
|
| 54 |
+
```python
|
| 55 |
+
text = "[insert text here]"
|
| 56 |
+
inputs = tokenizer(text,return_tensors='pt')
|
| 57 |
+
output = model(**inputs)
|
| 58 |
+
```
|