Spaces:
Runtime error
Runtime error
File size: 3,693 Bytes
1f8de22 a5e7dd0 1f8de22 a5e7dd0 1f8de22 72a0576 1f8de22 59c16a6 72a0576 1f8de22 4d84659 1f8de22 72a0576 1f8de22 fe0532d 1f8de22 eb844ee 1f8de22 b326cea 938f076 1f8de22 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 |
import torch
import torch.nn as nn
import gradio as gr
import numpy as np
import os
import random
from transformers import AutoConfig, AutoModel, AutoTokenizer
device = torch.device('cpu')
labels = {
0: 'toxic',
1: 'severe_toxic',
2: 'obscene',
3: 'threat',
4: 'insult',
5: 'identity_hate',
}
MODEL_NAME='roberta-base'
NUM_CLASSES=6
MAX_LEN = 128
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
class ToxicModel(torch.nn.Module):
def __init__(self):
super(ToxicModel, self).__init__()
hidden_dropout_prob: float = 0.1
layer_norm_eps: float = 1e-7
config = AutoConfig.from_pretrained(MODEL_NAME)
config.update(
{
"output_hidden_states": True,
"hidden_dropout_prob": hidden_dropout_prob,
"layer_norm_eps": layer_norm_eps,
"add_pooling_layer": False,
"num_labels": NUM_CLASSES,
}
)
self.transformer = AutoModel.from_pretrained(MODEL_NAME, config=config)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
self.dropout1 = nn.Dropout(0.1)
self.dropout2 = nn.Dropout(0.2)
self.dropout3 = nn.Dropout(0.3)
self.dropout4 = nn.Dropout(0.4)
self.dropout5 = nn.Dropout(0.5)
self.output = nn.Linear(config.hidden_size, NUM_CLASSES)
def forward(self, input_ids, attention_mask, token_type_ids):
transformer_out = self.transformer(input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids)
sequence_output = transformer_out[0]
sequence_output = self.dropout(torch.mean(sequence_output, 1))
logits1 = self.output(self.dropout1(sequence_output))
logits2 = self.output(self.dropout2(sequence_output))
logits3 = self.output(self.dropout3(sequence_output))
logits4 = self.output(self.dropout4(sequence_output))
logits5 = self.output(self.dropout5(sequence_output))
logits = (logits1 + logits2 + logits3 + logits4 + logits5) / 5
return logits
def inference_fn(model, input_ids=None, attention_mask=None, token_type_ids=None):
model.eval()
input_ids = input_ids[0].to(device)
attention_mask = attention_mask[0].to(device)
token_type_ids = token_type_ids[0].to(device)
with torch.no_grad():
output = model(input_ids.unsqueeze(0), attention_mask.unsqueeze(0), token_type_ids.unsqueeze(0))
out = output.sigmoid().detach().cpu().numpy().flatten()
return out
def predict(comment=None) -> dict:
text = str(comment)
text = " ".join(text.split())
inputs = tokenizer.encode_plus(
text,
None,
add_special_tokens=True,
max_length=MAX_LEN,
pad_to_max_length=True,
return_token_type_ids=True
)
ids = inputs['input_ids']
mask = inputs['attention_mask']
token_type_ids = inputs["token_type_ids"]
ids = torch.tensor(ids, dtype=torch.long),
mask = torch.tensor(mask, dtype=torch.long),
token_type_ids = torch.tensor(token_type_ids, dtype=torch.long),
model = ToxicModel()
model.load_state_dict(torch.load("toxicx_model_0.pth", map_location=torch.device(device)))
model.to(device)
predicted = inference_fn(model, ids, mask, token_type_ids)
return {labels[i]: float(predicted[i]) for i in range(NUM_CLASSES)}
gr.Interface(fn=predict,
inputs=gr.inputs.Textbox(lines=2, placeholder="Your Comment… "),
title="Toxic Comment Classification",
outputs=gr.outputs.Label(num_top_classes=NUM_CLASSES)).launch() |