qwen3-8b-base-beta-dpo-hh-harmless-4xh200-batch-64-20260424-025105

• Loss: 0.7256
• Beta Dpo/gap Mean: 9.9202
• Beta Dpo/gap Std: 18.3470
• Beta Dpo/beta Used Raw: 0.1809
• Beta Dpo/beta Used: 0.1995
• Beta Dpo/mask Keep Frac: 1.0
• Logits/chosen: 1.5449
• Logits/rejected: 1.4137

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 5e-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: 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

Framework versions

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