learning_rate: 0.0002
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
num_epochs: 5
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss
1.3937	0.2401	100	1.4147
1.3475	0.4802	200	1.3459
1.2985	0.7203	300	1.3032
1.2532	0.9604	400	1.2672
1.0762	1.1993	500	1.2640
1.0907	1.4394	600	1.2509
1.0635	1.6795	700	1.2279
1.0412	1.9196	800	1.2202
0.8275	2.1585	900	1.2694
0.8208	2.3986	1000	1.2789
0.8361	2.6387	1100	1.2626
0.8094	2.8788	1200	1.2588
0.5687	3.1176	1300	1.3845
0.5854	3.3577	1400	1.3959
0.5728	3.5978	1500	1.3886
0.5658	3.8379	1600	1.3923
0.4255	4.0768	1700	1.5321
0.4261	4.3169	1800	1.5281
0.4317	4.5570	1900	1.5352
0.4409	4.7971	2000	1.5324

MinhalZafar/uslegalqa-llama-3.2-3b-v2

uslegalqa-llama-3.2-3b-v2