distilbert-base-sarcasm

Fine-tuned distilbert-base-uncased on the BESSTIE-CW-26 dataset for binary sarcasm classification.

Training

• Base model: distilbert-base-uncased
• Task: sarcasm (binary)
• Epochs: 2
• Batch size: 4
• Learning rate: 2e-5
• Weight decay: 0.01
• Max sequence length: 64
• Seed: 42 (best of {42, 65, 131})
• Optimizer: AdamW (Trainer default)

Test results

• macro-F1: 0.6335

Usage

from transformers import AutoModelForSequenceClassification, AutoTokenizer

model = AutoModelForSequenceClassification.from_pretrained("vyshnav112233/distilbert-base-sarcasm")
tokenizer = AutoTokenizer.from_pretrained("vyshnav112233/distilbert-base-sarcasm")
inputs = tokenizer("your sentence here", return_tensors="pt", truncation=True, max_length=64)
logits = model(**inputs).logits