This model is a fine-tuned version of tsavage68/chat_600STEPS_1e8rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.6928
Rewards/chosen: 0.0004
Rewards/rejected: -0.0004
Rewards/accuracies: 0.4154
Rewards/margins: 0.0008
Logps/rejected: -18.8060
Logps/chosen: -16.7510
Logits/rejected: -0.5976
Logits/chosen: -0.5974

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: 1e-08
train_batch_size: 4
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 100
training_steps: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6933	0.0977	50	0.6938	-0.0003	0.0009	0.3956	-0.0012	-18.7927	-16.7574	-0.5974	-0.5973
0.6926	0.1953	100	0.6938	-0.0011	0.0002	0.3890	-0.0013	-18.7999	-16.7655	-0.5978	-0.5977
0.6939	0.2930	150	0.6934	0.0001	0.0006	0.3780	-0.0004	-18.7964	-16.7533	-0.5977	-0.5976
0.6935	0.3906	200	0.6933	-0.0003	-0.0002	0.4198	-0.0002	-18.8035	-16.7577	-0.5978	-0.5977
0.6934	0.4883	250	0.6931	0.0005	0.0004	0.4242	0.0001	-18.7978	-16.7493	-0.5980	-0.5979
0.6925	0.5859	300	0.6938	0.0000	0.0012	0.3912	-0.0012	-18.7895	-16.7546	-0.5982	-0.5981
0.6919	0.6836	350	0.6930	0.0002	-0.0002	0.4044	0.0004	-18.8044	-16.7530	-0.5977	-0.5976
0.6941	0.7812	400	0.6932	0.0001	0.0002	0.4154	-0.0001	-18.7997	-16.7532	-0.5979	-0.5978
0.6918	0.8789	450	0.6933	0.0005	0.0007	0.4154	-0.0002	-18.7949	-16.7500	-0.5982	-0.5981
0.694	0.9766	500	0.6935	-0.0000	0.0006	0.4132	-0.0006	-18.7963	-16.7548	-0.5978	-0.5977
0.692	1.0742	550	0.6934	0.0006	0.0011	0.4088	-0.0005	-18.7914	-16.7489	-0.5970	-0.5969
0.6927	1.1719	600	0.6935	0.0001	0.0008	0.3978	-0.0007	-18.7940	-16.7538	-0.5973	-0.5972
0.6942	1.2695	650	0.6931	0.0003	0.0001	0.4286	0.0002	-18.8007	-16.7513	-0.5974	-0.5973
0.6928	1.3672	700	0.6938	-0.0001	0.0012	0.3846	-0.0013	-18.7896	-16.7554	-0.5974	-0.5973
0.6934	1.4648	750	0.6938	-0.0002	0.0011	0.3934	-0.0013	-18.7914	-16.7566	-0.5976	-0.5975
0.6946	1.5625	800	0.6931	-0.0004	-0.0005	0.4132	0.0001	-18.8067	-16.7586	-0.5976	-0.5975
0.6928	1.6602	850	0.6928	0.0004	-0.0004	0.4154	0.0008	-18.8060	-16.7510	-0.5976	-0.5974
0.6924	1.7578	900	0.6928	0.0004	-0.0004	0.4154	0.0008	-18.8060	-16.7510	-0.5976	-0.5974
0.6939	1.8555	950	0.6928	0.0004	-0.0004	0.4154	0.0008	-18.8060	-16.7510	-0.5976	-0.5974
0.6928	1.9531	1000	0.6928	0.0004	-0.0004	0.4154	0.0008	-18.8060	-16.7510	-0.5976	-0.5974

Framework versions

Transformers 4.40.1
Pytorch 2.0.0+cu117
Datasets 2.19.1
Tokenizers 0.19.1

tsavage68/chat_1000_STEPS_01beta_1e8rate_CDPOSFT

chat_1000_STEPS_01beta_1e8rate_CDPOSFT

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
chat_1000_STEPS_01beta_1e8rate_CDPOSFT