genv3pair1NoGT_1.5B_cdpo_ebs32_lr5e-07_beta0.1_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: 0.3713
• Rewards/chosen: 0.9176
• Rewards/rejected: 0.0
• Rewards/accuracies: 0.9750
• Rewards/margins: 0.9176
• Logps/rejected: -32.3351
• Logps/chosen: -20.9565
• Logits/rejected: -3.1652
• Logits/chosen: -3.1781

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
0.6516	0.1117	20	0.6414	0.1186	0.0	1.0	0.1186	-40.4957	-28.9464	-2.3009	-2.4389
0.5135	0.2235	40	0.5086	0.4264	0.0	1.0	0.4264	-37.0924	-25.8683	-2.5744	-2.6771
0.4234	0.3352	60	0.4188	0.7095	0.0	0.9750	0.7095	-34.1697	-23.0372	-2.9217	-2.9733
0.3961	0.4469	80	0.3969	0.8053	0.0	0.9750	0.8053	-33.4004	-22.0792	-3.0136	-3.0527
0.3154	0.5587	100	0.3815	0.8689	0.0	0.9750	0.8689	-32.7767	-21.4436	-3.0932	-3.1168
0.3652	0.6704	120	0.3740	0.9063	0.0	0.9750	0.9063	-32.5288	-21.0689	-3.1456	-3.1596
0.3976	0.7821	140	0.3722	0.9151	0.0	0.9750	0.9151	-32.4026	-20.9812	-3.1588	-3.1709
0.3637	0.8939	160	0.3716	0.9131	0.0	0.9750	0.9131	-32.3961	-21.0012	-3.1560	-3.1672

Framework versions

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