This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.1 on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 1.0579
Rewards/chosen: -6.2250
Rewards/rejected: -6.0633
Rewards/accuracies: 0.4000
Rewards/margins: -0.1616
Logps/rejected: -77.5130
Logps/chosen: -77.5026
Logits/rejected: -4.8045
Logits/chosen: -4.8043

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: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 4
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 100
training_steps: 2000

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
1.2927	0.1	100	1.2512	-5.1921	-5.1652	0.4418	-0.0269	-68.5312	-67.1737	-2.1461	-2.1461
1.7128	0.2	200	1.4813	-7.0324	-6.9155	0.4308	-0.1169	-86.0349	-85.5773	-3.9413	-3.9413
1.5461	0.29	300	1.2731	-5.4304	-5.3353	0.4330	-0.0951	-70.2329	-69.5573	-3.1316	-3.1316
1.9939	0.39	400	1.2127	-5.0685	-4.9672	0.4505	-0.1013	-66.5519	-65.9385	-3.9101	-3.9101
1.5849	0.49	500	1.2395	-5.1346	-5.0482	0.4396	-0.0864	-67.3612	-66.5990	-3.6011	-3.6011
1.0981	0.59	600	1.2043	-4.9745	-4.8822	0.4440	-0.0923	-65.7019	-64.9985	-3.7103	-3.7103
1.9697	0.68	700	1.2507	-5.1232	-5.1033	0.4681	-0.0199	-67.9127	-66.4848	-3.6280	-3.6281
0.8747	0.78	800	1.1611	-4.8863	-4.7797	0.4505	-0.1065	-64.6770	-64.1157	-4.1453	-4.1453
1.004	0.88	900	1.1843	-5.6291	-5.5168	0.4527	-0.1123	-72.0471	-71.5438	-4.7506	-4.7510
1.1444	0.98	1000	1.2340	-6.0675	-6.1117	0.4571	0.0442	-77.9961	-75.9278	-4.5307	-4.5307
0.9495	1.07	1100	1.1048	-6.4880	-6.3179	0.3956	-0.1701	-80.0584	-80.1334	-4.5289	-4.5289
0.8455	1.17	1200	1.2109	-8.1849	-7.8918	0.3890	-0.2931	-95.7973	-97.1021	-5.1408	-5.1408
0.7447	1.27	1300	1.2187	-7.3352	-7.0426	0.3890	-0.2926	-87.3051	-88.6049	-4.4742	-4.4743
1.1554	1.37	1400	1.0728	-6.1506	-5.9622	0.3956	-0.1884	-76.5017	-76.7589	-5.0027	-5.0028
0.7376	1.47	1500	1.0798	-6.2916	-6.1208	0.4066	-0.1707	-78.0880	-78.1689	-4.9008	-4.9008
0.7962	1.56	1600	1.0666	-6.4071	-6.2400	0.4022	-0.1671	-79.2797	-79.3242	-4.8474	-4.8473
0.89	1.66	1700	1.0615	-6.3087	-6.1398	0.3912	-0.1690	-78.2774	-78.3405	-4.8210	-4.8209
0.902	1.76	1800	1.0582	-6.2322	-6.0708	0.4022	-0.1614	-77.5875	-77.5753	-4.7973	-4.7972
0.7122	1.86	1900	1.0578	-6.2254	-6.0638	0.4000	-0.1616	-77.5177	-77.5069	-4.8049	-4.8048
0.9455	1.95	2000	1.0579	-6.2250	-6.0633	0.4000	-0.1616	-77.5130	-77.5026	-4.8045	-4.8043

Framework versions

Transformers 4.39.1
Pytorch 2.0.0+cu117
Datasets 2.18.0
Tokenizers 0.15.2

tsavage68/v1_2000_STEPS_5e6_rate_01_beta_DPO

v1_2000_STEPS_5e6_rate_01_beta_DPO

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
v1_2000_STEPS_5e6_rate_01_beta_DPO