UTI2_M2_75steps_1e7rate_01beta_CSFTDPO

This model is a fine-tuned version of tsavage68/UTI_M2_1000steps_1e7rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1276
• Rewards/chosen: 0.0336
• Rewards/rejected: -4.2997
• Rewards/accuracies: 0.8800
• Rewards/margins: 4.3333
• Logps/rejected: -82.3535
• Logps/chosen: -19.5858
• Logits/rejected: -2.5678
• Logits/chosen: -2.5670

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: 75

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6904	0.3333	25	0.6532	0.0086	-0.0741	0.8500	0.0827	-40.0972	-19.8359	-2.6814	-2.6788
0.4057	0.6667	50	0.3414	0.0792	-0.9298	0.8800	1.0089	-48.6537	-19.1297	-2.6601	-2.6575
0.0995	1.0	75	0.1276	0.0336	-4.2997	0.8800	4.3333	-82.3535	-19.5858	-2.5678	-2.5670

Framework versions

• Transformers 4.41.2
• Pytorch 2.0.0+cu117
• Datasets 2.19.2
• Tokenizers 0.19.1