zephyr-7b-dpo-full-magpi-low-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: 0.0003
• Rewards/chosen: -2.0463
• Rewards/rejected: -52.8204
• Rewards/accuracies: 1.0
• Rewards/margins: 50.7741
• Logps/rejected: -5922.8276
• Logps/chosen: -571.6150
• Logits/rejected: -2.5402
• Logits/chosen: -3.0711

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	0.0024	-0.8703	-36.7088	0.9980	35.8385	-4311.6689	-454.0150	-2.5580	-2.8265
0.0061	0.2841	100	0.0013	-1.6291	-41.8173	0.9980	40.1883	-4822.5229	-529.8893	-2.6517	-2.9615
0.0019	0.4261	150	0.0005	-1.6961	-51.4828	1.0	49.7867	-5789.0674	-536.5906	-2.5182	-2.9830
0.0022	0.5682	200	0.0005	-2.0332	-50.2237	1.0	48.1905	-5663.1602	-570.3004	-2.5170	-2.9938
0.0066	0.7102	250	0.0007	-2.0455	-51.8463	1.0	49.8009	-5825.4209	-571.5287	-2.5713	-3.0874
0.0055	0.8523	300	0.0003	-2.0503	-53.0959	1.0	51.0456	-5950.3833	-572.0145	-2.5242	-3.0607
0.0011	0.9943	350	0.0003	-2.0463	-52.8204	1.0	50.7741	-5922.8276	-571.6150	-2.5402	-3.0711

Framework versions

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