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 new-dpo generated_from_trainer	HuggingFaceH4/ultrafeedback_binarized	name results llama3-8b-base-new-method-s_star0.6-20260426-230653

llama3-8b-base-new-method-s_star0.6-20260426-230653

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: 0.5352
• Fcm Dpo/beta: 0.0111
• Margin Dpo/margin Mean: 54.0836
• Margin Dpo/margin Std: 78.0441
• Logps/chosen: -383.4114
• Logps/rejected: -416.5982
• Logps/ref Chosen: -287.8268
• Logps/ref Rejected: -266.9300
• Logits/chosen: -0.8275
• Logits/rejected: -0.8179

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: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_batch_size: 8
• 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	Fcm Dpo/beta	Margin Dpo/margin Mean	Margin Dpo/margin Std	Logps/chosen	Logps/rejected	Logps/ref Chosen	Logps/ref Rejected	Logits/chosen	Logits/rejected
4.6147	0.4188	200	0.5500	0.0189	29.6010	45.7477	-320.2021	-328.9064	-287.8268	-266.9300	-0.8684	-0.8583
4.3361	0.8377	400	0.5352	0.0111	54.0836	78.0441	-383.4114	-416.5982	-287.8268	-266.9300	-0.8275	-0.8179

Framework versions

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