Model Card for zakariajaadi/distilbert-base-uncased-imdb

• This distilbert-base-uncased model was fine-tuned for sequence classification using the imdb dataset with Lora (Low-Rank Adaptation) to enable efficient fine-tuning while reducing the number of trainable parameters.
• The model was fine-tuned for 5 epochs with:
  • batch size: 16
  • learning rate: 5e-05
  • maximum sequence length: 256.
• The best validation accuracy the model achieved was 89.7%
• The evaluation set was created by sampling 15% of the training data.

Model settings and hyperparams

• max_length=256
• batch_size = 16
• num_epochs = 5
• lr = 5e-5
• warmup_ratio=0.1
• weight_decay=0.1

Lora settings

• r=4
• lora_alpha=32
• lora_dropout=0.01

📌 GitHub profile

For more projects and code, check out my GitHub: zakariajaadi