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

Model description

More information needed

Intended uses & limitations

More information needed

More information needed

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 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
0.9953	0.1512	100	0.6485	0.0613	9.1290	0.7724	1.1312	-0.4201	0.1227	11.2317	17.3662	-0.0262	0.0613	1.0	-3.3065	-3.3155
1.0952	0.3023	200	0.6746	0.0160	14.8251	0.3434	0.4801	-0.4659	0.0456	21.0985	27.0643	-0.2764	0.0160	1.0	-3.2893	-3.2944
0.957	0.4535	300	0.6778	0.0089	18.5486	0.2473	0.3133	-0.4818	0.0249	27.7148	34.2356	-0.4500	0.0089	1.0	-3.1462	-3.1521
1.3632	0.6047	400	0.7275	0.0294	21.1309	0.9101	1.1957	-0.4574	0.0566	28.7664	38.3785	-0.3581	0.0294	1.0	-3.0744	-3.0811
1.2415	0.7559	500	0.8583	0.0607	35.3740	2.9334	3.5974	-0.4353	0.0818	44.1570	58.8353	-0.4270	0.0607	1.0	-2.7623	-2.7786
2.3016	0.9070	600	1.2926	0.1601	39.2063	8.1315	10.1723	-0.3823	0.1665	43.6826	65.1157	-0.1686	0.1601	1.0	-2.7393	-2.7541