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

Loss: 0.2391
Rewards/chosen: -3.1721
Rewards/rejected: -8.7679
Rewards/accuracies: 0.8788
Rewards/margins: 5.5958
Logps/rejected: -1280.5232
Logps/chosen: -709.6791
Logits/rejected: 2.9862
Logits/chosen: 0.4871
Excess Loss: 0.0302
Alpha 0 Uf: 0.2677
Alpha 1 Rlced Conifer: 0.7323
Rewards/chosen 1 Rlced Conifer: -3.3519
Rewards/rejected 1 Rlced Conifer: -10.1355
Rewards/accuracies 1 Rlced Conifer: 0.9088
Rewards/margins 1 Rlced Conifer: 6.7836
Logps/rejected 1 Rlced Conifer: -1461.0847
Logps/chosen 1 Rlced Conifer: -758.7692
Logits/rejected 1 Rlced Conifer: 2.9834
Logits/chosen 1 Rlced Conifer: 0.2872
Task Loss 1 Rlced Conifer: 0.1744
Task Excess Loss 1 Rlced Conifer: 0.0378
Rewards/chosen 0 Uf: -2.5137
Rewards/rejected 0 Uf: -3.9578
Rewards/accuracies 0 Uf: 0.7751
Rewards/margins 0 Uf: 1.4442
Logps/rejected 0 Uf: -637.3895
Logps/chosen 0 Uf: -540.6270
Logits/rejected 0 Uf: 3.2024
Logits/chosen 0 Uf: 1.0821
Task Loss 0 Uf: 0.5033
Task Excess Loss 0 Uf: 0.0690

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: 8
gradient_accumulation_steps: 4
total_train_batch_size: 256
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: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Excess Loss	Alpha 0 Uf	Alpha 1 Rlced Conifer	Rewards/chosen 1 Rlced Conifer	Rewards/rejected 1 Rlced Conifer	Rewards/accuracies 1 Rlced Conifer	Rewards/margins 1 Rlced Conifer	Logps/rejected 1 Rlced Conifer	Logps/chosen 1 Rlced Conifer	Logits/rejected 1 Rlced Conifer	Logits/chosen 1 Rlced Conifer	Task Loss 1 Rlced Conifer	Task Excess Loss 1 Rlced Conifer	Rewards/chosen 0 Uf	Rewards/rejected 0 Uf	Rewards/accuracies 0 Uf	Rewards/margins 0 Uf	Logps/rejected 0 Uf	Logps/chosen 0 Uf	Logits/rejected 0 Uf	Logits/chosen 0 Uf	Task Loss 0 Uf	Task Excess Loss 0 Uf
0.1882	0.4997	360	0.2996	-1.6886	-4.1417	0.8609	2.4532	-817.9084	-561.3260	1.4584	0.3084	0.0858	0.8164	0.1836	-1.7447	-4.6283	0.8926	2.8836	-910.3677	-598.0539	1.2471	0.1366	0.2441	0.1077	-1.4688	-2.4264	0.7375	0.9576	-484.2446	-436.1386	2.3554	0.8449	0.5159	0.0745
0.1534	0.9993	720	0.2788	-1.6895	-4.6113	0.8656	2.9218	-864.8680	-561.4199	1.5835	0.1282	0.0703	0.8639	0.1361	-1.7298	-5.1653	0.8921	3.4355	-964.0696	-596.5645	1.3282	-0.0899	0.2304	0.0945	-1.5189	-2.6316	0.7670	1.1128	-504.7690	-441.1461	2.6475	0.8112	0.4886	0.0496
0.0947	1.4990	1080	0.2421	-2.6372	-7.6503	0.8797	5.0132	-1168.7697	-656.1883	2.9592	0.5518	0.0336	0.2372	0.7628	-2.7432	-8.7916	0.9108	6.0484	-1326.6932	-697.9009	2.9155	0.3378	0.1806	0.0448	-2.2397	-3.6057	0.7721	1.3660	-602.1759	-513.2244	3.3160	1.1969	0.4985	0.0623
0.0894	1.9986	1440	0.2391	-3.1721	-8.7679	0.8788	5.5958	-1280.5232	-709.6791	2.9862	0.4871	0.0302	0.2677	0.7323	-3.3519	-10.1355	0.9088	6.7836	-1461.0847	-758.7692	2.9834	0.2872	0.1744	0.0378	-2.5137	-3.9578	0.7751	1.4442	-637.3895	-540.6270	3.2024	1.0821	0.5033	0.0690

Framework versions

Transformers 4.44.2
Pytorch 2.2.0a0+81ea7a4
Datasets 2.21.0
Tokenizers 0.19.1

NicholasCorrado/zephyr-7b-uf-rlced-conifer-group-dpo-2e-alr-0.1

zephyr-7b-uf-rlced-conifer-group-dpo-2e-alr-0.1

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

NicholasCorrado/
zephyr-7b-uf-rlced-conifer-group-dpo-2e-alr-0.1