qlora

This model is a fine-tuned version of cardiffnlp/twitter-roberta-base-sentiment-latest on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1844
• Accuracy: 0.9322
• F1 Macro: 0.9322
• F1 Weighted: 0.9322

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 16
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.PAGED_ADAMW with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 50
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1 Macro	F1 Weighted
2.2579	0.1826	20	0.4075	0.8133	0.8128	0.8128
1.6504	0.3653	40	0.3658	0.8367	0.8350	0.8350
1.2766	0.5479	60	0.2479	0.8967	0.8966	0.8966
1.1452	0.7306	80	0.2252	0.9133	0.9133	0.9133
1.0420	0.9132	100	0.2207	0.9156	0.9154	0.9154
0.9200	1.0913	120	0.2104	0.9167	0.9167	0.9167
0.9177	1.2740	140	0.2089	0.9222	0.9222	0.9222
0.7950	1.4566	160	0.1962	0.9267	0.9266	0.9266
0.8560	1.6393	180	0.2120	0.9178	0.9176	0.9176
0.9545	1.8219	200	0.1914	0.9278	0.9278	0.9278
0.7653	2.0	220	0.1997	0.93	0.9299	0.9299
0.8118	2.1826	240	0.1898	0.9289	0.9289	0.9289
0.7282	2.3653	260	0.1910	0.9267	0.9266	0.9266
0.8910	2.5479	280	0.1836	0.9344	0.9344	0.9344
0.6892	2.7306	300	0.1842	0.9311	0.9311	0.9311
0.7994	2.9132	320	0.1843	0.9344	0.9344	0.9344
0.7994	3.0	330	0.1844	0.9322	0.9322	0.9322

Framework versions

• PEFT 0.18.1
• Transformers 5.4.0
• Pytorch 2.10.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2