library_name, base_model, tags, datasets, model-index

library_name	base_model	tags	datasets	model-index
transformers	mistral-7b-base-sft-hh-helpful-4xh200-batch-64-20260418-015332	alignment-handbook beta-dpo generated_from_trainer	Anthropic/hh-rlhf	name results mistral-7b-base-beta-dpo-hh-helpful-4xh200-batch-64-20260418-015332

mistral-7b-base-beta-dpo-hh-helpful-4xh200-batch-64-20260418-015332

This model is a fine-tuned version of mistral-7b-base-sft-hh-helpful-4xh200-batch-64-20260418-015332 on the Anthropic/hh-rlhf dataset. It achieves the following results on the evaluation set:

• Loss: 0.6015
• Beta Dpo/beta: 0.0010
• Beta Dpo/loss Margin Mean: 243.4043
• Beta Dpo/beta Margin Mean: 0.2434
• Beta Dpo/beta Margin Std: 0.4217
• Beta Dpo/beta Margin Grad Mean: -0.4422
• Beta Dpo/beta Margin Grad Std: 0.0983
• Beta Dpo/gap Mean: 404.4037
• Beta Dpo/gap Std: 357.4069
• Beta Dpo/beta Used Raw: -9.5600
• Beta Dpo/beta Used: 0.0010
• Beta Dpo/mask Keep Frac: 1.0
• Logits/chosen: -2.7813
• Logits/rejected: -2.8108

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: 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

Training Loss	Epoch	Step	Validation Loss	Beta Dpo/beta	Beta Dpo/loss Margin Mean	Beta Dpo/beta Margin Mean	Beta Dpo/beta Margin Std	Beta Dpo/beta Margin Grad Mean	Beta Dpo/beta Margin Grad Std	Beta Dpo/gap Mean	Beta Dpo/gap Std	Beta Dpo/beta Used Raw	Beta Dpo/beta Used	Beta Dpo/mask Keep Frac	Logits/chosen	Logits/rejected
1.3346	0.1468	100	0.7825	0.0211	38.6966	1.4685	2.0475	-0.4727	0.0403	60.6513	63.8526	-1.2173	0.0211	1.0	-2.9129	-2.9033
1.265	0.2937	200	1.2116	0.0416	108.9061	8.0746	10.4591	-0.4594	0.0608	175.9197	183.7102	-3.9208	0.0416	1.0	-2.3116	-2.3059
0.5857	0.4405	300	0.6708	0.0032	165.3890	0.8039	1.0106	-0.4553	0.0715	284.4015	265.4041	-7.0408	0.0032	1.0	-2.3951	-2.3756
3.7878	0.5874	400	0.6122	0.0010	205.4126	0.2054	0.3571	-0.4506	0.0845	362.1024	333.2912	-9.3014	0.0010	1.0	-2.4431	-2.4332
6.7444	0.7342	500	0.6026	0.0010	233.9227	0.2339	0.3910	-0.4441	0.0919	390.5113	345.8571	-9.2953	0.0010	1.0	-2.6421	-2.6564
0.5388	0.8811	600	0.6015	0.0010	243.4043	0.2434	0.4217	-0.4422	0.0983	404.4037	357.4069	-9.5600	0.0010	1.0	-2.7813	-2.8108

Framework versions

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