zephyr-dpop-qlora-uf-ours-5e-7-epoch1

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the generation/UF dataset. It achieves the following results on the evaluation set:

• Loss: 0.7024
• Positive Losses: 0.1617
• Dpo Losses: 0.6844
• Rewards/chosen: 0.0533
• Rewards/rejected: 0.0348
• Rewards/accuracies: 0.5940
• Rewards/margins: 0.0185
• Rewards/margins Max: 0.1082
• Rewards/margins Min: -0.0600
• Rewards/margins Std: 0.0561
• Logps/rejected: -255.0996
• Logps/chosen: -279.2664
• Logits/rejected: -2.7455
• Logits/chosen: -2.7839

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: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• total_eval_batch_size: 16
• 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	Positive Losses	Dpo Losses	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Rewards/margins Max	Rewards/margins Min	Rewards/margins Std	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6834	0.28	100	0.6932	0.0200	0.6905	0.0248	0.0194	0.5930	0.0054	0.0341	-0.0197	0.0178	-256.6423	-282.1181	-2.7657	-2.8044
0.6629	0.56	200	0.6977	0.1042	0.6860	0.0485	0.0335	0.5980	0.0149	0.0881	-0.0489	0.0456	-255.2263	-279.7464	-2.7492	-2.7879
0.6479	0.85	300	0.7024	0.1592	0.6845	0.0528	0.0346	0.5940	0.0182	0.1072	-0.0591	0.0555	-255.1217	-279.3168	-2.7478	-2.7859

Framework versions

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