Nvidia-Llama-3.1-Nemotron-Nano-4B-v1.1-Summarization-QLoRA

This model is a fine-tuned version of nvidia/Llama-3.1-Nemotron-Nano-4B-v1.1 on the scitldr dataset. It achieves the following results on the evaluation set:

• Loss: 2.6638

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: 1
• eval_batch_size: 1
• seed: 42
• 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: linear
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.7819	0.1004	200	2.7248
2.6369	0.2008	400	2.7038
2.6191	0.3012	600	2.6867
2.5851	0.4016	800	2.6784
2.5811	0.5020	1000	2.6674
2.59	0.6024	1200	2.6563
2.5887	0.7028	1400	2.6476
2.55	0.8032	1600	2.6456
2.5912	0.9036	1800	2.6312
2.5467	1.0040	2000	2.6378
2.1808	1.1044	2200	2.6606
2.2174	1.2048	2400	2.6817
2.1764	1.3052	2600	2.6726
2.2018	1.4056	2800	2.6818
2.1532	1.5060	3000	2.6793
2.1691	1.6064	3200	2.6659
2.1174	1.7068	3400	2.6727
2.1496	1.8072	3600	2.6726
2.1834	1.9076	3800	2.6638

Framework versions

• PEFT 0.16.0
• Transformers 4.54.0
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.2