train_winogrande_123_1760637730

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the winogrande dataset. It achieves the following results on the evaluation set:

• Loss: 0.0460
• Num Input Tokens Seen: 38394016

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: 5e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 123
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Input Tokens Seen
0.078	1.0	9090	0.0558	1918144
0.0031	2.0	18180	0.0485	3838192
0.005	3.0	27270	0.0460	5757648
0.0001	4.0	36360	0.0765	7676976
0.0004	5.0	45450	0.0684	9596496
0.0002	6.0	54540	0.0614	11516256
0.021	7.0	63630	0.0624	13435600
0.0	8.0	72720	0.0940	15356752
0.0001	9.0	81810	0.0731	17276752
0.0001	10.0	90900	0.0658	19196064
0.0	11.0	99990	0.0679	21115472
0.0	12.0	109080	0.0851	23035440
0.0	13.0	118170	0.0954	24955600
0.0	14.0	127260	0.0944	26875344
0.0	15.0	136350	0.1117	28795600
0.0	16.0	145440	0.1385	30715008
0.0	17.0	154530	0.1482	32634912
0.0	18.0	163620	0.1532	34554080
0.0	19.0	172710	0.1559	36472448
0.0	20.0	181800	0.1552	38394016

Framework versions

• PEFT 0.17.1
• Transformers 4.51.3
• Pytorch 2.9.0+cu128
• Datasets 4.0.0
• Tokenizers 0.21.4