learning_rate: 2e-05
train_batch_size: 1
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 2
gradient_accumulation_steps: 16
total_train_batch_size: 32
total_eval_batch_size: 2
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
lr_scheduler_warmup_ratio: 0.1
num_epochs: 3.0

Training Loss	Epoch	Step	Validation Loss
0.0056	0.1698	92	0.0098
0.0115	0.3397	184	0.0107
0.015	0.5095	276	0.0094
0.0082	0.6794	368	0.0104
0.0094	0.8492	460	0.0095
0.0038	1.0185	552	0.0086
0.0029	1.1883	644	0.0095
0.01	1.3581	736	0.0082
0.0019	1.5280	828	0.0081
0.0045	1.6978	920	0.0080
0.0091	1.8677	1012	0.0077
0.0057	2.0369	1104	0.0081
0.0006	2.2068	1196	0.0086
0.0075	2.3766	1288	0.0088
0.0065	2.5464	1380	0.0087
0.0084	2.7163	1472	0.0087
0.0027	2.8861	1564	0.0087

boradorish/qwen3-4b-base-prompt

sft_base