library_name, base_model, tags, datasets, model-index

library_name	base_model	tags	datasets	model-index
transformers	W-61/llama-3-8b-base-sft-ultrachat-8xh200	alignment-handbook ipo generated_from_trainer	HuggingFaceH4/ultrafeedback_binarized	name results llama-3-8b-base-ipo-ultrafeedback-4xh200-batch-128

llama-3-8b-base-ipo-ultrafeedback-4xh200-batch-128

This model is a fine-tuned version of W-61/llama-3-8b-base-sft-ultrachat-8xh200 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 2344.3516
• Rewards/chosen: -0.0325
• Rewards/rejected: -0.0526
• Rewards/accuracies: 0.6880
• Rewards/margins: 0.0202
• Logps/rejected: -6.5599
• Logps/chosen: -4.3618
• Logits/rejected: -0.4831
• Logits/chosen: -0.4853

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: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_batch_size: 16
• 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

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
19287.1	0.4188	200	2426.4592	-0.0183	-0.0264	0.6280	0.0081	-3.9353	-2.9447	-0.6160	-0.6205
18745.8219	0.8377	400	2344.3516	-0.0325	-0.0526	0.6880	0.0202	-6.5599	-4.3618	-0.4831	-0.4853

Framework versions

• Transformers 4.51.0
• Pytorch 2.3.1+cu121
• Datasets 2.21.0
• Tokenizers 0.21.4