SimpleTransformers Multi-label Text Classification Model loaded using HuggingFace Pipeline giving wrong output

[DrRaja]

Hi, I trained a Multi-Label Text Classification Model with distilbert-base-uncased. The model works as intended when I use it through the SimpleTransfomers wrapper. However, when I upload the same model to the HuggingFace Hub to consume it using the HuggingFace Pipeline, it only returns a single label and that is also completely incorrect. For example; a sentence like "I love you, I like you" returns the label "toxic" with a score of 0.7, whereas using my local model this is 0. Any idea what am I doing wrong?

The code to train the model:

from simpletransformers.classification import MultiLabelClassificationModel

model = MultiLabelClassificationModel('distilbert', 'distilbert-base-uncased', num_labels=6, args={'train_batch_size':2, 'gradient_accumulation_steps':16, 'learning_rate': 3e-5, 'num_train_epochs': 3, 'max_seq_length': 512})

and the code I'm using to generate the output using HuggingFace:

from transformers import AutoModelForSequenceClassification, AutoTokenizer, TextClassificationPipeline, pipeline

tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
model = AutoModelForSequenceClassification.from_pretrained(model_path)

classifier =  pipeline("text-classification", model=model, tokenizer=tokenizer)

classifier('i love you, i like you')

#output [{'label': 'toxic', 'score': 0.708755612373352}]

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.