zephyr-7b-dpo-full-magpi-high-curriculum

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

• Loss: 0.0002
• Rewards/chosen: -1.4221
• Rewards/rejected: -46.8461
• Rewards/accuracies: 1.0
• Rewards/margins: 45.4241
• Logps/rejected: -5325.4023
• Logps/chosen: -509.1871
• Logits/rejected: 1.3860
• Logits/chosen: -1.6588

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
0.0053	0.1420	50	0.0027	-0.5772	-33.4452	0.9980	32.8680	-3985.3066	-424.7043	-2.6572	-2.9202
0.0021	0.2841	100	0.0006	-1.4249	-49.2914	1.0	47.8664	-5569.9243	-509.4729	-2.1667	-3.0557
0.0023	0.4261	150	0.0004	-1.6397	-49.1673	1.0	47.5276	-5557.5210	-530.9537	-1.3032	-2.7968
0.0006	0.5682	200	0.0004	-1.3019	-43.7051	1.0	42.4032	-5011.2983	-497.1734	0.0532	-2.5388
0.0007	0.7102	250	0.0002	-1.4460	-47.2582	1.0	45.8122	-5366.6079	-511.5773	0.9592	-1.9357
0.0009	0.8523	300	0.0002	-1.4275	-46.7636	1.0	45.3360	-5317.1460	-509.7358	1.3477	-1.6957
0.0042	0.9943	350	0.0002	-1.4221	-46.8461	1.0	45.4241	-5325.4023	-509.1871	1.3860	-1.6588

Framework versions

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