zephyr-dpop-qlora-gpt4-5e-7

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the generation/GPT4 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7751
• Positive Losses: 0.8596
• Dpo Losses: 0.6847
• Rewards/chosen: 0.0247
• Rewards/rejected: 0.0066
• Rewards/accuracies: 0.5900
• Rewards/margins: 0.0182
• Rewards/margins Max: 0.1239
• Rewards/margins Min: -0.0706
• Rewards/margins Std: 0.0644
• Logps/rejected: -257.9209
• Logps/chosen: -282.1187
• Logits/rejected: -2.7289
• Logits/chosen: -2.7668

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
0.6709	0.28	100	0.7044	0.1334	0.6900	0.0150	0.0085	0.5830	0.0065	0.0434	-0.0241	0.0224	-257.7293	-283.0980	-2.7607	-2.7991
0.6375	0.56	200	0.7539	0.6398	0.6862	0.0226	0.0079	0.5880	0.0147	0.1022	-0.0596	0.0535	-257.7878	-282.3363	-2.7335	-2.7718
0.6203	0.85	300	0.7727	0.8307	0.6847	0.0252	0.0072	0.5920	0.0180	0.1227	-0.0701	0.0638	-257.8545	-282.0691	-2.7315	-2.7691

Framework versions

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