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: 0.6625
Rewards/chosen: -0.4073
Rewards/rejected: -0.5192
Rewards/accuracies: 0.5077
Rewards/margins: 0.1119
Logps/rejected: -23.9940
Logps/chosen: -20.8276
Logits/rejected: -0.8077
Logits/chosen: -0.8075

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-06
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
0.6925	0.0977	50	0.6917	0.0117	0.0085	0.4659	0.0031	-18.7166	-16.6380	-0.6015	-0.6013
0.6776	0.1953	100	0.6812	-0.0371	-0.0646	0.5253	0.0275	-19.4479	-17.1259	-0.6242	-0.6241
0.6915	0.2930	150	0.6811	-0.0700	-0.1026	0.4945	0.0325	-19.8275	-17.4549	-0.6198	-0.6196
0.6953	0.3906	200	0.6773	-0.0987	-0.1411	0.5209	0.0424	-20.2134	-17.7417	-0.6034	-0.6031
0.6947	0.4883	250	0.6764	-0.0437	-0.0935	0.5143	0.0498	-19.7369	-17.1915	-0.5932	-0.5930
0.6739	0.5859	300	0.6712	-0.1173	-0.1856	0.5077	0.0682	-20.6575	-17.9280	-0.6422	-0.6420
0.6648	0.6836	350	0.6740	-0.1818	-0.2444	0.5187	0.0626	-21.2460	-18.5724	-0.6126	-0.6123
0.6576	0.7812	400	0.6680	-0.1865	-0.2647	0.5143	0.0782	-21.4489	-18.6195	-0.6568	-0.6566
0.6645	0.8789	450	0.6702	-0.1619	-0.2377	0.5077	0.0759	-21.1794	-18.3732	-0.6161	-0.6159
0.6531	0.9766	500	0.6707	-0.2045	-0.2840	0.4989	0.0795	-21.6418	-18.7995	-0.6466	-0.6464
0.4839	1.0742	550	0.6670	-0.2362	-0.3265	0.5209	0.0903	-22.0674	-19.1168	-0.6941	-0.6939
0.4713	1.1719	600	0.6664	-0.3234	-0.4195	0.5209	0.0961	-22.9967	-19.9882	-0.7392	-0.7389
0.4965	1.2695	650	0.6644	-0.3536	-0.4578	0.5121	0.1043	-23.3802	-20.2903	-0.7732	-0.7730
0.4232	1.3672	700	0.6632	-0.3744	-0.4833	0.5187	0.1088	-23.6347	-20.4989	-0.7902	-0.7899
0.4192	1.4648	750	0.6627	-0.3915	-0.5018	0.5143	0.1103	-23.8201	-20.6699	-0.8012	-0.8009
0.4946	1.5625	800	0.6629	-0.3997	-0.5103	0.5187	0.1106	-23.9054	-20.7517	-0.8051	-0.8049
0.4538	1.6602	850	0.6630	-0.4061	-0.5168	0.5143	0.1107	-23.9701	-20.8155	-0.8072	-0.8069
0.4873	1.7578	900	0.6630	-0.4074	-0.5183	0.5121	0.1109	-23.9850	-20.8283	-0.8080	-0.8078
0.4596	1.8555	950	0.6628	-0.4070	-0.5181	0.5143	0.1112	-23.9834	-20.8241	-0.8078	-0.8075
0.4361	1.9531	1000	0.6625	-0.4073	-0.5192	0.5077	0.1119	-23.9940	-20.8276	-0.8077	-0.8075

Framework versions

Transformers 4.40.1
Pytorch 2.0.0+cu117
Datasets 2.19.0
Tokenizers 0.19.1

tsavage68/chat_1000_STEPS_01beta_1e6_rate_CDPOSFT

chat_1000_STEPS_01beta_1e6_rate_CDPOSFT

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
chat_1000_STEPS_01beta_1e6_rate_CDPOSFT