phi-2-distilabeled-intel-orca-dpo

This model is a fine-tuned version of microsoft/phi-2 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3007
• Rewards/chosen: 0.0679
• Rewards/rejected: -1.3058
• Rewards/accuracies: 0.9265
• Rewards/margins: 1.3737
• Logps/rejected: -265.9995
• Logps/chosen: -117.6325
• Logits/rejected: 0.3789
• Logits/chosen: 0.3129

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: 1e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 20
• training_steps: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6803	0.06	20	0.6533	0.0100	-0.0741	0.8386	0.0841	-253.6825	-118.2115	0.3760	0.3395
0.5855	0.12	40	0.5133	0.0376	-0.3927	0.9195	0.4303	-256.8685	-117.9349	0.3829	0.3363
0.452	0.19	60	0.3846	0.0620	-0.8319	0.9246	0.8938	-261.2602	-117.6917	0.3875	0.3290
0.3538	0.25	80	0.3200	0.0716	-1.1689	0.9269	1.2405	-264.6308	-117.5958	0.3836	0.3189
0.295	0.31	100	0.3007	0.0679	-1.3058	0.9265	1.3737	-265.9995	-117.6325	0.3789	0.3129

Framework versions

• PEFT 0.7.1
• Transformers 4.37.1
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1