rlced-conifer-zephyr-7b-dpo-2e

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the data/rlced_conifer dataset. It achieves the following results on the evaluation set:

• Loss: 0.1578
• Rewards/chosen: -8.8806
• Rewards/rejected: -21.7173
• Rewards/accuracies: 0.9216
• Rewards/margins: 12.8367
• Logps/rejected: -2617.6685
• Logps/chosen: -1311.4659
• Logits/rejected: 3.0195
• Logits/chosen: -0.1677

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: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.1586	0.4982	240	0.1689	-2.6868	-8.8615	0.9252	6.1746	-1332.0890	-692.0928	1.5975	-0.1975
0.1408	0.9964	480	0.1482	-6.5979	-15.9265	0.9326	9.3286	-2038.5934	-1083.2030	3.3424	0.5015
0.0852	1.4946	720	0.1644	-10.0141	-23.7065	0.9228	13.6924	-2816.5886	-1424.8193	3.4873	0.0636
0.0743	1.9927	960	0.1578	-8.8688	-21.7065	0.9203	12.8376	-2616.5852	-1310.2896	3.0165	-0.1724

Framework versions

• Transformers 4.44.1
• Pytorch 2.1.2+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1