genv3pair1NoGT_1.5B_cdpo_lm1_ebs32_lr5e-07_beta0.4_epoch1.0_42

This model is a fine-tuned version of Qwen/Qwen2.5-1.5B-Instruct on the YuchenLi01/MATH_Qwen2.5-1.5BInstruct_DPO_MoreUniqueResponseNoGTv3pair1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0981
• Rewards/chosen: 3.0826
• Rewards/rejected: 0.0
• Rewards/accuracies: 0.9750
• Rewards/margins: 3.0826
• Logps/rejected: -33.6270
• Logps/chosen: -22.3636
• Logits/rejected: -2.9597
• Logits/chosen: -3.0085

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-07
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5278	0.1117	20	0.5088	0.4050	0.0	1.0	0.4050	-40.5040	-29.0576	-2.3030	-2.4408
0.215	0.2235	40	0.2036	1.6290	0.0	1.0	1.6290	-37.2403	-25.9975	-2.5596	-2.6605
0.1164	0.3352	60	0.1178	2.6736	0.0	1.0	2.6736	-34.3945	-23.3861	-2.8632	-2.9246
0.0929	0.4469	80	0.1082	2.8582	0.0	0.9750	2.8582	-34.0888	-22.9245	-2.9190	-2.9753
0.0561	0.5587	100	0.1016	2.9526	0.0	0.9750	2.9526	-33.8596	-22.6885	-2.9319	-2.9854
0.0772	0.6704	120	0.0972	3.0433	0.0	1.0	3.0433	-33.7543	-22.4618	-2.9473	-2.9971
0.1301	0.7821	140	0.0965	3.0222	0.0	1.0	3.0222	-33.5886	-22.5146	-2.9612	-3.0115
0.0991	0.8939	160	0.0970	3.0411	0.0	0.9750	3.0411	-33.6275	-22.4672	-2.9579	-3.0071

Framework versions

• Transformers 4.45.2
• Pytorch 2.5.1+cu121
• Datasets 3.5.0
• Tokenizers 0.20.3