zephyr-dpo-timedial-selfgen

This model is a fine-tuned version of EllieS/zephyr-sft-timedial on the EllieS/timedial_selfgen dataset. It achieves the following results on the evaluation set:

• Loss: 0.0055
• Rewards/chosen: 0.4843
• Rewards/rejected: -5.0618
• Rewards/accuracies: 1.0
• Rewards/margins: 5.5461
• Logps/rejected: -740.9361
• Logps/chosen: -96.9992
• Logits/rejected: -3.1155
• Logits/chosen: -2.6815

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-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• 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
0.0166	0.35	100	0.0198	0.4885	-3.4443	1.0	3.9328	-579.1801	-96.5815	-3.1058	-2.6808
0.0015	0.7	200	0.0055	0.4843	-5.0618	1.0	5.5461	-740.9361	-96.9992	-3.1155	-2.6815

Framework versions

• PEFT 0.7.1
• Transformers 4.39.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.2