mistral-7b-dpo-full-sft-wo-kqa_silver_wogold

This model is a fine-tuned version of Minbyul/mistral-7b-wo-kqa_silver_wogold-sft on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.0530
• Rewards/chosen: -2.4760
• Rewards/rejected: -21.0723
• Rewards/accuracies: 0.9700
• Rewards/margins: 18.5963
• Logps/rejected: -2709.2131
• Logps/chosen: -407.7003
• Logits/rejected: -2.0225
• Logits/chosen: -2.2276

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: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• 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
0.2735	0.32	100	0.0529	-1.3592	-8.1857	0.9700	6.8265	-1420.5509	-296.0260	-2.7457	-2.5375
0.1321	0.63	200	0.0507	-2.0405	-16.8511	0.9600	14.8106	-2287.0967	-364.1557	-2.2518	-2.3349
0.117	0.95	300	0.0531	-2.4855	-21.1345	0.9700	18.6490	-2715.4331	-408.6504	-2.0210	-2.2273

Framework versions

• Transformers 4.39.0.dev0
• Pytorch 2.1.2
• Datasets 2.14.6
• Tokenizers 0.15.2