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: 0.6934
Rewards/chosen: 0.0042
Rewards/rejected: 0.0031
Rewards/accuracies: 0.4176
Rewards/margins: 0.0011
Logps/rejected: -18.7959
Logps/chosen: -16.7463
Logits/rejected: -0.5975
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
0.6957	0.0977	50	0.6960	0.0006	0.0049	0.3648	-0.0043	-18.7923	-16.7534	-0.5974	-0.5973
0.69	0.1953	100	0.6940	0.0010	0.0011	0.4308	-0.0001	-18.7998	-16.7526	-0.5985	-0.5983
0.7001	0.2930	150	0.6953	0.0032	0.0058	0.4132	-0.0026	-18.7904	-16.7482	-0.5979	-0.5978
0.6948	0.3906	200	0.6936	-0.0016	-0.0025	0.4352	0.0009	-18.8070	-16.7578	-0.5978	-0.5977
0.6978	0.4883	250	0.6946	0.0002	0.0016	0.4198	-0.0014	-18.7989	-16.7543	-0.5980	-0.5978
0.6903	0.5859	300	0.6944	0.0032	0.0040	0.4088	-0.0008	-18.7940	-16.7482	-0.5970	-0.5969
0.6902	0.6836	350	0.6956	-0.0040	-0.0007	0.3560	-0.0033	-18.8034	-16.7625	-0.5971	-0.5970
0.6966	0.7812	400	0.6948	-0.0005	0.0011	0.4198	-0.0016	-18.7999	-16.7557	-0.5978	-0.5976
0.6891	0.8789	450	0.6932	0.0019	0.0000	0.4308	0.0019	-18.8019	-16.7508	-0.5973	-0.5972
0.6907	0.9766	500	0.6940	0.0028	0.0027	0.4352	0.0001	-18.7966	-16.7490	-0.5974	-0.5973
0.6941	1.0742	550	0.6916	0.0010	-0.0039	0.4330	0.0049	-18.8098	-16.7526	-0.5974	-0.5972
0.6857	1.1719	600	0.6949	0.0012	0.0030	0.4132	-0.0018	-18.7960	-16.7523	-0.5975	-0.5973
0.6927	1.2695	650	0.6926	0.0050	0.0020	0.4308	0.0029	-18.7979	-16.7447	-0.5977	-0.5976
0.6844	1.3672	700	0.6954	-0.0029	-0.0002	0.4044	-0.0027	-18.8024	-16.7604	-0.5978	-0.5977
0.6951	1.4648	750	0.6940	0.0000	0.0000	0.4176	-0.0000	-18.8020	-16.7546	-0.5978	-0.5977
0.6965	1.5625	800	0.6947	0.0046	0.0060	0.4132	-0.0014	-18.7900	-16.7455	-0.5975	-0.5974
0.6894	1.6602	850	0.6934	0.0043	0.0031	0.4176	0.0012	-18.7958	-16.7461	-0.5976	-0.5974
0.6882	1.7578	900	0.6934	0.0042	0.0031	0.4176	0.0011	-18.7959	-16.7463	-0.5975	-0.5974
0.6932	1.8555	950	0.6934	0.0042	0.0031	0.4176	0.0011	-18.7959	-16.7463	-0.5975	-0.5974
0.6899	1.9531	1000	0.6934	0.0042	0.0031	0.4176	0.0011	-18.7959	-16.7463	-0.5975	-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_05beta_1e8rate_CDPOSFT

chat_1000_STEPS_05beta_1e8rate_CDPOSFT

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
chat_1000_STEPS_05beta_1e8rate_CDPOSFT