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: 0.6923
Rewards/chosen: -0.0014
Rewards/rejected: -0.0031
Rewards/accuracies: 0.4352
Rewards/margins: 0.0018
Logps/rejected: -18.8334
Logps/chosen: -16.7684
Logits/rejected: -0.5994
Logits/chosen: -0.5993

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: 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
0.6944	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
0.6937	0.2930	150	0.6929	-0.0008	-0.0013	0.4242	0.0005	-18.8152	-16.7631	-0.5980	-0.5979
0.6909	0.3906	200	0.6929	-0.0011	-0.0016	0.4110	0.0005	-18.8177	-16.7654	-0.5980	-0.5979
0.6939	0.4883	250	0.6925	-0.0009	-0.0022	0.4527	0.0013	-18.8240	-16.7635	-0.5982	-0.5981
0.6914	0.5859	300	0.6925	-0.0020	-0.0035	0.4308	0.0014	-18.8366	-16.7748	-0.5990	-0.5989
0.6922	0.6836	350	0.6926	-0.0031	-0.0043	0.4527	0.0012	-18.8453	-16.7857	-0.5985	-0.5984
0.6926	0.7812	400	0.6924	-0.0021	-0.0036	0.4440	0.0015	-18.8380	-16.7757	-0.5992	-0.5991
0.6912	0.8789	450	0.6922	-0.0021	-0.0041	0.4615	0.0021	-18.8432	-16.7752	-0.5984	-0.5982
0.6918	0.9766	500	0.6921	-0.0018	-0.0040	0.4418	0.0022	-18.8422	-16.7723	-0.5986	-0.5985
0.69	1.0742	550	0.6918	-0.0017	-0.0045	0.4637	0.0028	-18.8469	-16.7718	-0.5988	-0.5987
0.6882	1.1719	600	0.6923	-0.0013	-0.0031	0.4659	0.0018	-18.8330	-16.7675	-0.5994	-0.5993
0.6887	1.2695	650	0.6924	-0.0019	-0.0036	0.4308	0.0016	-18.8375	-16.7739	-0.5988	-0.5987
0.6886	1.3672	700	0.6918	-0.0003	-0.0030	0.4549	0.0028	-18.8325	-16.7572	-0.5991	-0.5989
0.6876	1.4648	750	0.6919	-0.0005	-0.0031	0.4725	0.0026	-18.8327	-16.7592	-0.5994	-0.5993
0.6921	1.5625	800	0.6914	-0.0001	-0.0038	0.4725	0.0037	-18.8396	-16.7556	-0.5994	-0.5992
0.6882	1.6602	850	0.6920	-0.0006	-0.0029	0.4945	0.0023	-18.8307	-16.7602	-0.5996	-0.5994
0.69	1.7578	900	0.6920	-0.0010	-0.0033	0.4505	0.0023	-18.8350	-16.7647	-0.5995	-0.5993
0.6888	1.8555	950	0.6923	-0.0014	-0.0032	0.4352	0.0018	-18.8340	-16.7686	-0.5994	-0.5993
0.6878	1.9531	1000	0.6923	-0.0014	-0.0031	0.4352	0.0018	-18.8334	-16.7684	-0.5994	-0.5993

Framework versions

Transformers 4.40.1
Pytorch 2.0.0+cu117
Datasets 2.19.1
Tokenizers 0.19.1

tsavage68/chat_1000_STEPS_01beta_1e7rate_CDPOSFT

chat_1000_STEPS_01beta_1e7rate_CDPOSFT

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
chat_1000_STEPS_01beta_1e7rate_CDPOSFT