dpo_from_multiple_samples_shortest_numina_aime

This model is a fine-tuned version of bespokelabs/Bespoke-Stratos-7B on the mlfoundations-dev/dpo_from_multiple_samples_shortest_numina_aime dataset. It achieves the following results on the evaluation set:

• Loss: 4.5934
• Rewards/chosen: -7.6229
• Rewards/rejected: -8.1168
• Rewards/accuracies: 0.5859
• Rewards/margins: 0.4940
• Logps/chosen: -0.7623
• Logps/rejected: -0.8117
• Logits/chosen: -0.6184
• Logits/rejected: -0.4483

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: 8e-07
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 32
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• total_eval_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/chosen	Logps/rejected	Logits/chosen	Logits/rejected
4.3711	0.96	18	4.5934	-7.6229	-8.1168	0.5859	0.4940	-0.7623	-0.8117	-0.6184	-0.4483

Framework versions

• Transformers 4.46.1
• Pytorch 2.3.0
• Datasets 3.1.0
• Tokenizers 0.20.3