zephyr-7b-ultra-p-0.03

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: 0.5093
• Rewards/chosen: -0.5738
• Rewards/rejected: -1.9426
• Rewards/accuracies: 0.7188
• Rewards/margins: 1.3688
• Logps/rejected: -266.9053
• Logps/chosen: -235.7460
• Logits/rejected: -2.5874
• Logits/chosen: -2.6524

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: 1
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5792	0.1030	100	0.5523	-0.2490	-0.8768	0.6953	0.6278	-256.2472	-232.4978	-2.5792	-2.6447
0.5414	0.2060	200	0.5345	-0.5674	-1.5354	0.7031	0.9680	-262.8332	-235.6819	-2.4884	-2.5509
0.5332	0.3090	300	0.5228	-0.4996	-1.3691	0.6797	0.8695	-261.1697	-235.0035	-2.5201	-2.5875
0.517	0.4120	400	0.5328	-0.6976	-2.1007	0.7344	1.4031	-268.4859	-236.9835	-2.6132	-2.6770
0.5108	0.5150	500	0.5212	-0.4373	-1.7338	0.6953	1.2966	-264.8175	-234.3803	-2.6314	-2.6952
0.5027	0.6180	600	0.5215	-0.4051	-1.7631	0.7266	1.3580	-265.1102	-234.0588	-2.5777	-2.6427
0.52	0.7210	700	0.5197	-0.4182	-1.6831	0.7188	1.2649	-264.3105	-234.1898	-2.6053	-2.6677
0.5308	0.8240	800	0.5167	-0.5398	-1.9222	0.7266	1.3824	-266.7009	-235.4056	-2.6061	-2.6709
0.4846	0.9270	900	0.5086	-0.5308	-1.8569	0.7109	1.3261	-266.0478	-235.3156	-2.5822	-2.6476

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.20.0