learning_rate: 5e-06
train_batch_size: 1
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 8
total_train_batch_size: 8
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: 0.05
num_epochs: 1.0

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/chosen	Logps/rejected	Logits/chosen	Logits/rejected
No log	0	0	0.6931	0.0	0.0	0.0	0.0	-175.0362	-170.8915	-1.4694	-1.1752
0.0009	0.2513	99	0.0020	-1.4484	-12.4536	1.0	11.0052	-189.5201	-295.4278	-1.8259	-1.5958
0.0001	0.5025	198	0.0006	-3.6703	-17.2659	1.0	13.5956	-211.7392	-343.5504	-1.9436	-1.7229
0.0015	0.7538	297	0.0006	-3.8635	-17.6150	1.0	13.7515	-213.6709	-347.0411	-1.9430	-1.7233

akshay-sked/qwen3-14b-svamp-full-dpo

qwen3-14b-svamp-final-base-to-full-purebf16-1xb200-dpo-cutoff4096-1epoch