This model is a fine-tuned version of Minbyul/selfbiorag-7b-wo-live_qa-sft on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

Loss: 0.1422
Rewards/chosen: -1.2709
Rewards/rejected: -13.2633
Rewards/accuracies: 0.9167
Rewards/margins: 11.9924
Logps/rejected: -1991.3534
Logps/chosen: -456.8682
Logits/rejected: -0.4049
Logits/chosen: -0.4878

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-06
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: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.2635	0.3	100	0.1990	-0.5114	-9.8179	0.9167	9.3065	-1646.8138	-380.9204	-0.1085	-0.3091
0.1415	0.61	200	0.1502	-0.9081	-11.0651	0.9167	10.1570	-1771.5302	-420.5836	-0.4280	-0.4824
0.0892	0.91	300	0.1421	-1.2604	-13.2286	0.9167	11.9683	-1987.8828	-455.8129	-0.4048	-0.4887

Framework versions

Transformers 4.39.0.dev0
Pytorch 2.1.2
Datasets 2.14.6
Tokenizers 0.15.2

Minbyul/selfbiorag-7b-dpo-full-sft-wo-live_qa

selfbiorag-7b-dpo-full-sft-wo-live_qa

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Minbyul/
selfbiorag-7b-dpo-full-sft-wo-live_qa