zephyr-7b-dpo-full-prometheus-low-curriculum

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4969
• Rewards/chosen: -1.0990
• Rewards/rejected: -1.9910
• Rewards/accuracies: 0.7328
• Rewards/margins: 0.8920
• Logps/rejected: -447.3745
• Logps/chosen: -369.8574
• Logits/rejected: 0.8388
• Logits/chosen: 0.0187

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: 55
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• 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: 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
0.6558	0.1143	50	0.6420	-0.0440	-0.1562	0.6724	0.1122	-263.8943	-264.3628	-2.5310	-2.5775
0.5602	0.2286	100	0.5737	-0.9542	-1.4308	0.6509	0.4765	-391.3545	-355.3837	-1.4641	-1.6090
0.533	0.3429	150	0.5293	-0.6925	-1.3308	0.6853	0.6382	-381.3522	-329.2125	-0.9684	-1.4379
0.5466	0.4571	200	0.5157	-0.7144	-1.4929	0.6983	0.7785	-397.5714	-331.4009	-0.1901	-0.8315
0.5044	0.5714	250	0.5078	-0.9462	-1.7789	0.6853	0.8327	-426.1694	-354.5842	0.4657	-0.2137
0.4967	0.6857	300	0.5044	-1.0772	-1.9084	0.7371	0.8312	-439.1140	-367.6776	0.4821	-0.2917
0.5055	0.8	350	0.4981	-1.0311	-1.9125	0.7414	0.8814	-439.5230	-363.0710	0.6801	-0.1449
0.5131	0.9143	400	0.4969	-1.0990	-1.9910	0.7328	0.8920	-447.3745	-369.8574	0.8388	0.0187

Framework versions

• Transformers 4.44.0.dev0
• Pytorch 2.1.2
• Datasets 2.20.0
• Tokenizers 0.19.1