This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the HuggingFaceH4/hh-rlhf-h4 dataset.
It achieves the following results on the evaluation set:
Loss: 0.5453
Rewards/chosen: -2.4749
Rewards/rejected: -3.2088
Rewards/accuracies: 0.7174
Rewards/margins: 0.7340
Logps/rejected: -474.4292
Logps/chosen: -394.6965
Logits/rejected: 3.9043
Logits/chosen: 3.4245
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: 8
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 4
total_train_batch_size: 256
total_eval_batch_size: 64
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
