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: 0.6924
Rewards/chosen: -0.0020
Rewards/rejected: -0.0034
Rewards/accuracies: 0.4505
Rewards/margins: 0.0015
Logps/rejected: -18.8364
Logps/chosen: -16.7742
Logits/rejected: -0.5980
Logits/chosen: -0.5979

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-07
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: 550

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6944	0.0977	50	0.6937	-0.0002	0.0007	0.3846	-0.0010	-18.7946	-16.7570	-0.5974	-0.5972
0.6929	0.1953	100	0.6932	-0.0013	-0.0013	0.4352	0.0000	-18.8149	-16.7673	-0.5987	-0.5985
0.693	0.2930	150	0.6929	-0.0015	-0.0021	0.4264	0.0006	-18.8229	-16.7696	-0.5983	-0.5982
0.6939	0.3906	200	0.6934	-0.0009	-0.0005	0.4000	-0.0004	-18.8074	-16.7637	-0.5983	-0.5982
0.6942	0.4883	250	0.6928	-0.0014	-0.0021	0.4352	0.0008	-18.8233	-16.7684	-0.5976	-0.5974
0.6928	0.5859	300	0.6929	-0.0025	-0.0032	0.4462	0.0006	-18.8338	-16.7801	-0.5984	-0.5983
0.6912	0.6836	350	0.6929	-0.0025	-0.0031	0.4198	0.0005	-18.8325	-16.7799	-0.5977	-0.5976
0.6928	0.7812	400	0.6926	-0.0019	-0.0030	0.4484	0.0011	-18.8320	-16.7736	-0.5986	-0.5984
0.6915	0.8789	450	0.6928	-0.0022	-0.0029	0.4396	0.0007	-18.8311	-16.7765	-0.5989	-0.5987
0.6925	0.9766	500	0.6928	-0.0024	-0.0032	0.4088	0.0008	-18.8341	-16.7791	-0.5980	-0.5978
0.6916	1.0742	550	0.6924	-0.0020	-0.0034	0.4505	0.0015	-18.8364	-16.7742	-0.5980	-0.5979

Framework versions

Transformers 4.40.1
Pytorch 2.0.0+cu117
Datasets 2.19.1
Tokenizers 0.19.1

tsavage68/chat_550_STEPS_01beta_1e7rate_CDPOSFT

chat_550_STEPS_01beta_1e7rate_CDPOSFT

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
chat_550_STEPS_01beta_1e7rate_CDPOSFT