zephyr-7b-dpo-full-magpi-reward-scale-01

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: 0.0001
• Rewards/chosen: -2.1581
• Rewards/rejected: -72.1363
• Rewards/accuracies: 1.0
• Rewards/margins: 69.9782
• Logps/rejected: -7854.4189
• Logps/chosen: -582.7947
• Logits/rejected: 2.3418
• Logits/chosen: -1.3156

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
0.0066	0.1420	50	0.0037	-1.1750	-45.6975	0.9960	44.5225	-5210.5396	-484.4839	-2.4201	-2.9075
0.0019	0.2841	100	0.0003	-1.8979	-62.8577	1.0	60.9597	-6926.5537	-556.7739	-0.2036	-2.7790
0.001	0.4261	150	0.0002	-1.8922	-67.8911	1.0	65.9989	-7429.9023	-556.2059	1.8870	-2.6554
0.0013	0.5682	200	0.0001	-2.1555	-73.7897	1.0	71.6342	-8019.7568	-582.5308	2.4298	-2.2775
0.0015	0.7102	250	0.0001	-2.1984	-72.3465	1.0	70.1481	-7875.4424	-586.8257	2.3430	-1.6277
0.0009	0.8523	300	0.0001	-2.1293	-71.2932	1.0	69.1638	-7770.1060	-579.9163	2.3229	-1.4073
0.0015	0.9943	350	0.0001	-2.1581	-72.1363	1.0	69.9782	-7854.4189	-582.7947	2.3418	-1.3156

Framework versions

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