tulu2-7b-cost-UF-UI-HHRLHF-2e-6

This model is a fine-tuned version of allenai/tulu-2-7b on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7862
• Rewards/chosen: -2.7944
• Rewards/rejected: -3.1878
• Rewards/accuracies: 0.5755
• Rewards/margins: 0.3934
• Rewards/margins Max: 3.4649
• Rewards/margins Min: -1.8768
• Rewards/margins Std: 1.7287
• Logps/rejected: -637.1887
• Logps/chosen: -616.4665
• Logits/rejected: 1.0155
• Logits/chosen: 0.8493

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: 2e-06
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• total_eval_batch_size: 16
• 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	Rewards/margins Max	Rewards/margins Min	Rewards/margins Std	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.198	1.0	3974	0.7862	-2.7944	-3.1878	0.5755	0.3934	3.4649	-1.8768	1.7287	-637.1887	-616.4665	1.0155	0.8493

Framework versions

• PEFT 0.7.1
• Transformers 4.39.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.2