zephyr-7b-dpo-full-ultrabin-amazon

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.6914
• Rewards/chosen: 0.0
• Rewards/rejected: 0.0
• Rewards/accuracies: 0.0
• Rewards/margins: 0.0
• Logps/rejected: -2352.0
• Logps/chosen: -2624.0
• Logits/rejected: -1.5781
• Logits/chosen: -1.5781

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: 55
• 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
0.6914	0.1050	50	0.6914	0.0	0.0	0.0	0.0	-3024.0	-3440.0	-2.4844	-2.5
0.6914	0.2101	100	0.6914	0.0	0.0	0.0	0.0	-2544.0	-2880.0	-2.0312	-2.0469
0.6914	0.3151	150	0.6914	0.0	0.0	0.0	0.0	-2432.0	-2736.0	-1.5938	-1.5859
0.6914	0.4202	200	0.6914	0.0	0.0	0.0	0.0	-2384.0	-2688.0	-1.6875	-1.6875
0.6914	0.5252	250	0.6914	0.0	0.0	0.0	0.0	-2352.0	-2640.0	-1.7031	-1.7031
0.6914	0.6303	300	0.6914	0.0	0.0	0.0	0.0	-2336.0	-2624.0	-1.6797	-1.6719
0.6914	0.7353	350	0.6914	0.0	0.0	0.0	0.0	-2352.0	-2640.0	-1.6016	-1.6016
0.6914	0.8403	400	0.6914	0.0	0.0	0.0	0.0	-2368.0	-2640.0	-1.5781	-1.5781
0.6914	0.9454	450	0.6914	0.0	0.0	0.0	0.0	-2352.0	-2624.0	-1.5781	-1.5781

Framework versions

• Transformers 4.44.0.dev0
• Pytorch 2.1.2
• Datasets 2.20.0
• Tokenizers 0.19.1