emotions-entailment/roberta-raw

roberta-emotion-raw

This model is a fine-tuned version of FacebookAI/roberta-base. It achieves the following results on the evaluation set:

• Loss: 0.3922
• F1 Micro: 0.8552
• F1 Macro: 0.8040
• Precision Micro: 0.8693
• Recall Micro: 0.8416

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 3
• total_train_batch_size: 24
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2095
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	F1 Macro	F1 Micro	Precision Micro	Recall Micro	Validation Loss
1.1898	1.0	10479	0.7956	0.8419	0.8473	0.8367	0.3967
1.1891	2.0	20958	0.8115	0.8554	0.8658	0.8454	0.3942

Framework versions

• Transformers 5.10.2
• Pytorch 2.11.0+cu128
• Datasets 4.8.5
• Tokenizers 0.22.2