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

Loss: 0.4588
Margin Dpo/margin Mean: 11.1187
Margin Dpo/margin Std: 15.0696
Logps/chosen: -119.7147
Logps/rejected: -113.9535
Logps/ref Chosen: -97.0617
Logps/ref Rejected: -80.1818
Logits/chosen: -0.5876
Logits/rejected: -0.5495

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: 16
eval_batch_size: 16
seed: 42
distributed_type: multi-GPU
num_devices: 8
total_train_batch_size: 128
total_eval_batch_size: 128
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	Margin Dpo/margin Mean	Margin Dpo/margin Std	Logps/chosen	Logps/rejected	Logps/ref Chosen	Logps/ref Rejected	Logits/chosen	Logits/rejected
0.4265	0.2941	100	0.5427	5.3418	10.0613	-107.1989	-95.6607	-97.0617	-80.1818	-0.6361	-0.6086
0.3411	0.5882	200	0.4754	10.2996	14.6526	-119.3164	-112.7360	-97.0617	-80.1818	-0.6021	-0.5640
0.3552	0.8824	300	0.4588	11.1187	15.0696	-119.7147	-113.9535	-97.0617	-80.1818	-0.5876	-0.5495

Framework versions

Transformers 4.51.0
Pytorch 2.3.1+cu121
Datasets 2.21.0
Tokenizers 0.21.4

W-61/llama-3-8b-base-margin-dpo-hh-helpful-8xh200

llama-3-8b-base-margin-dpo-hh-helpful-8xh200-20260410-172009

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

W-61/
llama-3-8b-base-margin-dpo-hh-helpful-8xh200