zephyr-dpop-qlora-uf6k-5e-6

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the generation/UF6konly dataset. It achieves the following results on the evaluation set:

• Loss: 0.6881
• Positive Losses: 0.0953
• Dpo Losses: 0.6680
• Rewards/chosen: 0.1485
• Rewards/rejected: 0.0945
• Rewards/accuracies: 0.6920
• Rewards/margins: 0.0540
• Rewards/margins Max: 0.2140
• Rewards/margins Min: -0.0946
• Rewards/margins Std: 0.1026
• Logps/rejected: -249.1260
• Logps/chosen: -269.7429
• Logits/rejected: -2.7470
• Logits/chosen: -2.7842

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: 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
0.6896	0.3	100	0.6927	0.0589	0.6767	0.1278	0.0930	0.6710	0.0348	0.1572	-0.0767	0.0771	-249.2807	-271.8154	-2.7452	-2.7829
0.6919	0.61	200	0.6864	0.0952	0.6689	0.1438	0.0918	0.6830	0.0521	0.2115	-0.0943	0.1012	-249.3993	-270.2086	-2.7481	-2.7856
0.6774	0.91	300	0.6879	0.0941	0.6679	0.1483	0.0942	0.6900	0.0541	0.2146	-0.0945	0.1028	-249.1596	-269.7678	-2.7464	-2.7837

Framework versions

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