Instructions to use debajyotimaz/codemix_hate with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use debajyotimaz/codemix_hate with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="debajyotimaz/codemix_hate")# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("debajyotimaz/codemix_hate") model = AutoModelForSequenceClassification.from_pretrained("debajyotimaz/codemix_hate", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Commit ·
a3800ec
1
Parent(s): 8f2eefc
second model version
Browse files
models.py
DELETED
|
@@ -1,50 +0,0 @@
|
|
| 1 |
-
import torch
|
| 2 |
-
from transformers import BertTokenizer
|
| 3 |
-
from torch import nn
|
| 4 |
-
from transformers import BertModel
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
#CustomMBERTModel= torch.load("/data2/Akash_for_interface/model_mbert_1416.pt")
|
| 8 |
-
tokenizer = BertTokenizer.from_pretrained('bert-base-multilingual-cased')
|
| 9 |
-
sentences= "you are good person."
|
| 10 |
-
max_len=len(sentences)
|
| 11 |
-
encoding = tokenizer.encode_plus(sentences,add_special_tokens=True,max_length=max_len,padding='max_length',truncation=True,return_tensors='pt')
|
| 12 |
-
input_ids=encoding['input_ids'].flatten(),
|
| 13 |
-
attention_mask= encoding['attention_mask'].flatten()
|
| 14 |
-
#print(input_ids[0])
|
| 15 |
-
|
| 16 |
-
labels=["Non Hateful","Hateful"]
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
device = 'cpu'
|
| 21 |
-
class CustomMBERTModel(nn.Module):
|
| 22 |
-
def __init__(self, num_labels):
|
| 23 |
-
super(CustomMBERTModel, self).__init__()
|
| 24 |
-
self.bert = BertModel.from_pretrained('bert-base-multilingual-cased')
|
| 25 |
-
|
| 26 |
-
# Freeze all layers except the top 2
|
| 27 |
-
for param in self.bert.parameters():
|
| 28 |
-
param.requires_grad = False
|
| 29 |
-
|
| 30 |
-
# Unfreeze the parameters of the top 2 layers
|
| 31 |
-
for param in self.bert.encoder.layer[-2:].parameters():
|
| 32 |
-
param.requires_grad = True
|
| 33 |
-
|
| 34 |
-
# Adding Linear layer
|
| 35 |
-
self.linear = nn.Linear(self.bert.config.hidden_size, num_labels)
|
| 36 |
-
|
| 37 |
-
def forward(self, input_ids, attention_mask):
|
| 38 |
-
outputs = self.bert(input_ids=input_ids, attention_mask=attention_mask)
|
| 39 |
-
hidden_states = outputs.last_hidden_state
|
| 40 |
-
|
| 41 |
-
pooled_output= hidden_states[:,0,:]
|
| 42 |
-
pooled_output=torch.squeeze(pooled_output,dim=1)
|
| 43 |
-
|
| 44 |
-
#print('p-shape:', pooled_output.shape)
|
| 45 |
-
|
| 46 |
-
logits = self.linear(pooled_output)
|
| 47 |
-
|
| 48 |
-
#print('l-shape:', logits.shape)
|
| 49 |
-
return logits
|
| 50 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|