DeepSeek-R1-Distill-Llama-8B-Summarization-QLoRa

This model is a fine-tuned version of deepseek-ai/DeepSeek-R1-Distill-Llama-8B on the scitldr dataset. It achieves the following results on the evaluation set:

• Loss: 2.5393

Model description

DeepSeek-R1-Distill-Llama-8B fine-tuned for summarization of scientific documents

Intended uses & limitations

Summarization

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: 2
• eval_batch_size: 2
• 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
• lr_scheduler_warmup_steps: 2
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.459	0.2209	220	2.4903
2.3971	0.4418	440	2.4720
2.3821	0.6627	660	2.4550
2.3665	0.8835	880	2.4392
2.3582	1.1044	1100	2.5203
1.7824	1.3253	1320	2.5360
1.7599	1.5462	1540	2.5486
1.7352	1.7671	1760	2.5404
1.7088	1.9880	1980	2.5393

Framework versions

• PEFT 0.14.0
• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0