spin-v-trans

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5983
• Rewards/real: -0.8207
• Rewards/generated: -1.5904
• Rewards/accuracies: 0.6831
• Rewards/margins: 0.7697
• Logps/generated: -132.1463
• Logps/real: -142.7898
• Logits/generated: -2.7538
• Logits/real: -2.8344

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: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/real	Rewards/generated	Rewards/accuracies	Rewards/margins	Logps/generated	Logps/real	Logits/generated	Logits/real
0.6035	0.06	100	0.5925	0.0221	-0.4614	0.6871	0.4834	-120.8558	-134.3616	-2.9112	-2.9634
0.6154	0.13	200	0.5946	-0.3013	-0.8642	0.6887	0.5629	-124.8842	-137.5959	-2.8464	-2.8927
0.7218	0.19	300	0.6093	0.4121	-0.1100	0.6791	0.5220	-117.3417	-130.4615	-2.8558	-2.9389
0.6415	0.26	400	0.6003	-0.0211	-0.6809	0.6943	0.6598	-123.0508	-134.7932	-2.8630	-2.9343
0.6876	0.32	500	0.6062	-0.7329	-1.3805	0.6728	0.6475	-130.0468	-141.9119	-2.8364	-2.9102
0.5849	0.38	600	0.6079	-1.0222	-1.8413	0.6831	0.8190	-134.6551	-144.8048	-2.8346	-2.9254
0.5948	0.45	700	0.6026	-0.3949	-1.0789	0.6815	0.6840	-127.0314	-138.5318	-2.8405	-2.9282
0.6723	0.51	800	0.6017	-1.0046	-1.7350	0.6791	0.7303	-133.5921	-144.6288	-2.8398	-2.9189
0.6436	0.58	900	0.6029	-0.8739	-1.5819	0.6688	0.7080	-132.0616	-143.3213	-2.8060	-2.8862
0.6163	0.64	1000	0.6010	-0.7903	-1.6415	0.6783	0.8512	-132.6574	-142.4857	-2.8003	-2.8780
0.6297	0.7	1100	0.6027	-0.8523	-1.5836	0.6688	0.7313	-132.0778	-143.1050	-2.7848	-2.8601
0.6468	0.77	1200	0.5993	-0.8984	-1.7091	0.6839	0.8107	-133.3330	-143.5663	-2.7862	-2.8608
0.5741	0.83	1300	0.6019	-0.8763	-1.6648	0.6871	0.7886	-132.8906	-143.3451	-2.7748	-2.8527
0.5778	0.9	1400	0.5983	-0.8584	-1.6390	0.6839	0.7805	-132.6320	-143.1669	-2.7596	-2.8410
0.5929	0.96	1500	0.5983	-0.8207	-1.5904	0.6831	0.7697	-132.1463	-142.7898	-2.7538	-2.8344

Framework versions

• Transformers 4.37.0
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.2