learning_rate: 0.0003
train_batch_size: 128
eval_batch_size: 16
seed: 42
distributed_type: multi-GPU
gradient_accumulation_steps: 4
total_train_batch_size: 512
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-07 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 1

Training Loss	Epoch	Step	Validation Loss
3.5926	0.0312	1000	0.3328
3.6234	0.0624	2000	0.3559
3.8785	0.0935	3000	0.3841
4.1961	0.1247	4000	0.3729
3.9691	0.1559	5000	0.3557
3.8045	0.1871	6000	0.3395
3.9398	0.2182	7000	0.3382
3.868	0.2494	8000	0.3255
3.5309	0.2806	9000	0.3232
3.3898	0.3118	10000	0.3103
2.943	0.3429	11000	0.3042
3.1098	0.3741	12000	0.3001
2.9047	0.4053	13000	0.2934
3.1891	0.4365	14000	0.2920
2.6195	0.4676	15000	0.2839
2.6316	0.4988	16000	0.2825
2.6707	0.5300	17000	0.2767
2.3398	0.5612	18000	0.2751
2.4426	0.5923	19000	0.2712
2.6562	0.6235	20000	0.2678
2.5717	0.6547	21000	0.2712
2.2953	0.6859	22000	0.2648
2.2137	0.7170	23000	0.2644
1.974	0.7482	24000	0.2652
1.9008	0.7794	25000	0.2670
2.1832	0.8106	26000	0.2658
1.9268	0.8417	27000	0.2674
1.7371	0.8729	28000	0.2673
1.7844	0.9041	29000	0.2664
2.3234	0.9353	30000	0.2660
2.0305	0.9664	31000	0.2664
1.9336	0.9976	32000	0.2665

ferrazzipietro/unsup-gemma-3-1b-it-datav2

unsup-gemma-3-1b-it-datav2