This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-qlora on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

Loss: 1.3540
Rewards/chosen: -0.0639
Rewards/rejected: -0.3680
Rewards/accuracies: 0.7050
Rewards/margins: 0.3041
Logps/rejected: -2.2845
Logps/chosen: -1.1386
Logits/rejected: -1.9700
Logits/chosen: -2.0510

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-06
train_batch_size: 2
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 5
gradient_accumulation_steps: 4
total_train_batch_size: 40
total_eval_batch_size: 20
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
1.3799	0.0654	100	1.3804	-0.0062	-0.0408	0.6600	0.0346	-1.1940	-0.9462	-2.1974	-2.2810
1.3728	0.1308	200	1.3734	-0.0308	-0.1119	0.6900	0.0811	-1.4310	-1.0283	-2.2618	-2.3330
1.3605	0.1963	300	1.3670	-0.0656	-0.2070	0.7200	0.1414	-1.7478	-1.1442	-2.1971	-2.2674
1.3607	0.2617	400	1.3637	-0.0644	-0.2551	0.6975	0.1908	-1.9084	-1.1401	-2.2602	-2.3277
1.3642	0.3271	500	1.3625	-0.0744	-0.3109	0.6875	0.2366	-2.0943	-1.1734	-2.1841	-2.2534
1.3489	0.3925	600	1.3649	-0.1095	-0.4197	0.6850	0.3101	-2.4568	-1.2906	-2.0263	-2.1039
1.3735	0.4580	700	1.3653	-0.1046	-0.4143	0.7000	0.3097	-2.4389	-1.2743	-1.9237	-2.0155
1.3606	0.5234	800	1.3592	-0.0745	-0.3701	0.6950	0.2956	-2.2915	-1.1739	-1.9493	-2.0356
1.3462	0.5888	900	1.3568	-0.0854	-0.3668	0.7050	0.2815	-2.2807	-1.2100	-1.9785	-2.0609
1.3527	0.6542	1000	1.3548	-0.0626	-0.3514	0.7050	0.2888	-2.2291	-1.1342	-1.9978	-2.0771
1.3483	0.7197	1100	1.3558	-0.0665	-0.3741	0.7025	0.3076	-2.3048	-1.1471	-1.9802	-2.0598
1.3558	0.7851	1200	1.3542	-0.0628	-0.3646	0.7050	0.3018	-2.2733	-1.1348	-1.9719	-2.0522
1.3515	0.8505	1300	1.3543	-0.0644	-0.3702	0.7050	0.3058	-2.2918	-1.1402	-1.9694	-2.0505
1.3572	0.9159	1400	1.3540	-0.0639	-0.3674	0.7075	0.3035	-2.2825	-1.1385	-1.9716	-2.0522
1.3527	0.9814	1500	1.3541	-0.0637	-0.3677	0.7025	0.3039	-2.2834	-1.1380	-1.9704	-2.0513

Framework versions

PEFT 0.10.0
Transformers 4.43.1
Pytorch 2.1.2+cu121
Datasets 2.18.0
Tokenizers 0.19.1

Kimory-X/zephyr-7b-mypo3_sim-qlora-lr5e6-beta0.30

zephyr-7b-mypo3_sim-qlora-lr5e6-beta0.30

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Kimory-X/
zephyr-7b-mypo3_sim-qlora-lr5e6-beta0.30