zephyr-7b-ultra-p-0.07

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: 0.4875
• Rewards/chosen: -0.5244
• Rewards/rejected: -1.7320
• Rewards/accuracies: 0.7109
• Rewards/margins: 1.2075
• Logps/rejected: -264.7986
• Logps/chosen: -235.2518
• Logits/rejected: -2.5806
• Logits/chosen: -2.6463

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
0.546	0.1030	100	0.5298	-0.2496	-1.0248	0.6875	0.7752	-257.7271	-232.5039	-2.5880	-2.6521
0.5221	0.2060	200	0.5055	-0.4087	-1.6295	0.6953	1.2208	-263.7740	-234.0950	-2.5117	-2.5716
0.5224	0.3090	300	0.4941	-0.4567	-1.5733	0.7031	1.1166	-263.2119	-234.5742	-2.5233	-2.5891
0.4907	0.4120	400	0.4953	-0.5910	-1.7671	0.75	1.1761	-265.1498	-235.9173	-2.5826	-2.6460
0.4876	0.5150	500	0.4960	-0.3941	-1.5280	0.6953	1.1339	-262.7591	-233.9482	-2.5836	-2.6465
0.4822	0.6180	600	0.4927	-0.3062	-1.5501	0.7109	1.2439	-262.9799	-233.0692	-2.5521	-2.6147
0.5079	0.7210	700	0.4948	-0.4099	-1.5277	0.7109	1.1178	-262.7565	-234.1069	-2.5957	-2.6577
0.5052	0.8240	800	0.4936	-0.4967	-1.7099	0.6953	1.2132	-264.5780	-234.9745	-2.5953	-2.6602
0.4691	0.9270	900	0.4879	-0.4878	-1.6605	0.7188	1.1727	-264.0843	-234.8857	-2.5774	-2.6430

Framework versions

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