zephyr-dpo-qlora-uf6k-5e-7

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: 0.6890
• Rewards/chosen: 0.0030
• Rewards/rejected: -0.0067
• Rewards/accuracies: 0.6750
• Rewards/margins: 0.0097
• Rewards/margins Max: 0.0448
• Rewards/margins Min: -0.0217
• Rewards/margins Std: 0.0219
• Logps/rejected: -259.2527
• Logps/chosen: -284.2960
• Logits/rejected: -2.7684
• Logits/chosen: -2.8066

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	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
0.6914	0.3	100	0.6915	-0.0004	-0.0041	0.6460	0.0037	0.0187	-0.0095	0.0093	-258.9901	-284.6319	-2.7677	-2.8065
0.6884	0.61	200	0.6895	0.0023	-0.0061	0.6850	0.0084	0.0389	-0.0189	0.0190	-259.1880	-284.3611	-2.7665	-2.8049
0.6873	0.91	300	0.6889	0.0031	-0.0066	0.6760	0.0098	0.0449	-0.0216	0.0219	-259.2438	-284.2815	-2.7640	-2.8026

Framework versions

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