train_winogrande_101112_1760638071

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: 0.0490
• Num Input Tokens Seen: 38366624

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: 101112
• 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
0.1336	1.0	9090	0.0617	1917952
0.0317	2.0	18180	0.0564	3835840
0.0003	3.0	27270	0.0654	5753152
0.0019	4.0	36360	0.0490	7672000
0.0008	5.0	45450	0.0590	9590080
0.0469	6.0	54540	0.0772	11509088
0.0004	7.0	63630	0.0771	13427712
0.0	8.0	72720	0.0949	15346672
0.0	9.0	81810	0.1089	17265344
0.0	10.0	90900	0.0961	19184224
0.0	11.0	99990	0.1043	21102912
0.0	12.0	109080	0.0872	23021312
0.0	13.0	118170	0.0864	24938688
0.0	14.0	127260	0.1193	26857088
0.0	15.0	136350	0.1188	28775840
0.0	16.0	145440	0.1330	30693088
0.0	17.0	154530	0.1429	32612480
0.0	18.0	163620	0.1453	34530176
0.0	19.0	172710	0.1471	36447600
0.0	20.0	181800	0.1468	38366624

Framework versions

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