midm-7B-nsmc

Model description

Intended uses & limitations

Training and evaluation data

• training data: nsmc의 train 데이터 중 상위 2000개
• evaluation data: nsmc의 test 데이터 중 상위 1000개

Training hyperparameters

• learning_rate: 0.0001
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 2
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.03
• training_steps: 300
• mixed_precision_training: Native AMP

Training results

• global_step=300
• training_loss=1.1095316060384115
• metrics={'train_runtime': 1012.8423, 'train_samples_per_second': 0.592,
  'train_steps_per_second': 0.296,
  'total_flos': 9315508499251200.0,
  'train_loss': 1.1095316060384115,
  'epoch': 0.3}

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0

Accuracy

• accuracy: 0.89