gemma-2-9b-it-dpo-1000

This model is a fine-tuned version of google/gemma-2-9b-it on the bct_non_cot_dpo_1000 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2974
• Rewards/chosen: -0.1483
• Rewards/rejected: -2.5189
• Rewards/accuracies: 0.8700
• Rewards/margins: 2.3706
• Logps/chosen: -31.5266
• Logps/rejected: -58.9056
• Logits/chosen: -8.2139
• Logits/rejected: -6.9867

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-06
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_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: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/chosen	Logps/rejected	Logits/chosen	Logits/rejected
0.5146	1.7778	50	0.4857	0.5450	-0.0018	0.8500	0.5468	-24.5940	-33.7344	-6.5014	-5.6575
0.3352	3.5556	100	0.3103	0.2930	-1.4831	0.8600	1.7760	-27.1142	-48.5476	-7.3532	-6.2995
0.2271	5.3333	150	0.3008	0.0199	-2.1688	0.8600	2.1887	-29.8448	-55.4048	-7.9154	-6.7490
0.2421	7.1111	200	0.2974	-0.1483	-2.5189	0.8700	2.3706	-31.5266	-58.9056	-8.2139	-6.9867
0.2241	8.8889	250	0.2987	-0.2014	-2.6345	0.8600	2.4332	-32.0576	-60.0622	-8.3179	-7.0668

Framework versions

• PEFT 0.12.0
• Transformers 4.45.2
• Pytorch 2.3.0
• Datasets 2.19.0
• Tokenizers 0.20.0