IE_L3_100steps_1e7rate_05beta_cSFTDPO

This model is a fine-tuned version of tsavage68/IE_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1803
• Rewards/chosen: -0.7835
• Rewards/rejected: -7.7662
• Rewards/accuracies: 0.7400
• Rewards/margins: 6.9827
• Logps/rejected: -91.1596
• Logps/chosen: -84.3647
• Logits/rejected: -0.7936
• Logits/chosen: -0.7251

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-07
• train_batch_size: 2
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• 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
0.4416	0.4	50	0.3457	-0.0969	-1.1506	0.7400	1.0537	-77.9284	-82.9916	-0.7954	-0.7373
0.1388	0.8	100	0.1803	-0.7835	-7.7662	0.7400	6.9827	-91.1596	-84.3647	-0.7936	-0.7251

Framework versions

• Transformers 4.44.2
• Pytorch 2.0.0+cu117
• Datasets 3.0.0
• Tokenizers 0.19.1