zephyr-7b-dpo-full

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5225
• Rewards/chosen: -0.5814
• Rewards/rejected: -1.3885
• Rewards/accuracies: 0.7773
• Rewards/margins: 0.8071
• Logps/rejected: -403.6498
• Logps/chosen: -323.4799
• Logits/rejected: 1.6879
• Logits/chosen: 1.2050

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: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• total_eval_batch_size: 64
• 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

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5999	0.2092	100	0.5966	-0.3739	-0.7450	0.7109	0.3711	-339.3055	-302.7284	-1.6763	-1.7949
0.5606	0.4184	200	0.5555	-0.6215	-1.2285	0.7461	0.6069	-387.6506	-327.4932	1.5409	1.1102
0.5211	0.6276	300	0.5284	-0.6372	-1.3932	0.7617	0.7559	-404.1211	-329.0649	1.2079	0.7847
0.5189	0.8368	400	0.5225	-0.5814	-1.3885	0.7773	0.8071	-403.6498	-323.4799	1.6879	1.2050

Framework versions

• Transformers 4.45.2
• Pytorch 2.1.0
• Datasets 3.0.1
• Tokenizers 0.20.1