easylm-ultrafeedback-sft-gemma-2-2b

This model is a fine-tuned version of google/gemma-2-2b on the ultrafeedback-sft dataset. It achieves the following results on the evaluation set:

• Loss: 1.2897

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: 3e-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 2
• total_eval_batch_size: 2
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
1.5578	0.0371	500	1.4651
1.4645	0.0742	1000	1.4362
1.4198	0.1113	1500	1.4196
1.3469	0.1484	2000	1.4051
1.3816	0.1855	2500	1.3920
1.3653	0.2226	3000	1.3809
1.4087	0.2596	3500	1.3715
1.2973	0.2967	4000	1.3615
1.348	0.3338	4500	1.3545
1.4639	0.3709	5000	1.3480
1.4405	0.4080	5500	1.3408
1.2926	0.4451	6000	1.3349
1.3452	0.4822	6500	1.3268
1.3076	0.5193	7000	1.3202
1.2696	0.5564	7500	1.3154
1.3833	0.5935	8000	1.3104
1.3217	0.6306	8500	1.3060
1.2351	0.6677	9000	1.3026
1.5295	0.7047	9500	1.2990
1.293	0.7418	10000	1.2967
1.2231	0.7789	10500	1.2942
1.2721	0.8160	11000	1.2926
1.3877	0.8531	11500	1.2913
1.2929	0.8902	12000	1.2903
1.4017	0.9273	12500	1.2900
1.2126	0.9644	13000	1.2897

Framework versions

• Transformers 4.43.3
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1