Mistral-Nemo-MCAI-SFT-DPO

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

Mistral-Nemo-Instruct-MCAI-SFT-DPO-3Epochs

This model is a fine-tuned version of pbevan11/Mistral-Nemo-Instruct-MCAI-SFT-3Epochs on the pbevan11/multilingual-constitutional-preference-pairs and the pbevan11/ultrafeedback_binarized_multilingual datasets. It achieves the following results on the evaluation set:

• Loss: 0.5579
• Rewards/chosen: -0.4264
• Rewards/rejected: -0.9042
• Rewards/accuracies: 0.7466
• Rewards/margins: 0.4777
• Logps/rejected: -213.7833
• Logps/chosen: -192.1762
• Logits/rejected: 0.5653
• Logits/chosen: 0.5394

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-07
• train_batch_size: 6
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 48
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0