zephyr-7b-align-scan-0.0-0.4-polynomial-2

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.8224
• Rewards/chosen: -0.8205
• Rewards/rejected: -1.9403
• Rewards/accuracies: 0.3433
• Rewards/margins: 1.1198
• Logps/rejected: -85.6178
• Logps/chosen: -76.3898
• Logits/rejected: -2.4242
• Logits/chosen: -2.4424

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: 8.772160997458124e-07
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: polynomial
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5775	1.0417	100	0.7712	0.7556	0.1333	0.3313	0.6223	-80.8200	-72.7430	-2.5200	-2.5356

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0
• Datasets 2.21.0
• Tokenizers 0.19.1