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: 0.6892
Rewards/chosen: -0.0034
Rewards/rejected: -0.0121
Rewards/accuracies: 0.4725
Rewards/margins: 0.0086
Logps/rejected: -18.8422
Logps/chosen: -16.7661
Logits/rejected: -0.5986
Logits/chosen: -0.5984

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
0.6925	0.0977	50	0.6943	0.0013	0.0029	0.3846	-0.0016	-18.7922	-16.7503	-0.5979	-0.5978
0.6919	0.1953	100	0.6932	-0.0001	-0.0007	0.4110	0.0005	-18.8042	-16.7551	-0.5986	-0.5985
0.6942	0.2930	150	0.6933	-0.0039	-0.0042	0.4176	0.0003	-18.8160	-16.7678	-0.5979	-0.5977
0.6964	0.3906	200	0.6932	-0.0035	-0.0040	0.4352	0.0005	-18.8154	-16.7662	-0.5984	-0.5983
0.6945	0.4883	250	0.6914	-0.0028	-0.0069	0.4505	0.0041	-18.8249	-16.7639	-0.5977	-0.5976
0.6906	0.5859	300	0.6920	-0.0066	-0.0096	0.4440	0.0031	-18.8341	-16.7765	-0.5985	-0.5984
0.6871	0.6836	350	0.6906	-0.0055	-0.0114	0.4440	0.0059	-18.8400	-16.7730	-0.5982	-0.5981
0.6889	0.7812	400	0.6897	-0.0066	-0.0143	0.4703	0.0076	-18.8495	-16.7768	-0.5990	-0.5989
0.689	0.8789	450	0.6905	-0.0053	-0.0115	0.4396	0.0063	-18.8404	-16.7722	-0.5986	-0.5984
0.6915	0.9766	500	0.6896	-0.0031	-0.0110	0.4681	0.0079	-18.8388	-16.7650	-0.5990	-0.5989
0.6834	1.0742	550	0.6906	-0.0031	-0.0091	0.4418	0.0060	-18.8323	-16.7650	-0.5987	-0.5986
0.683	1.1719	600	0.6894	-0.0041	-0.0125	0.4615	0.0084	-18.8437	-16.7683	-0.5991	-0.5990
0.6814	1.2695	650	0.6890	-0.0031	-0.0123	0.4681	0.0092	-18.8430	-16.7650	-0.5992	-0.5991
0.6811	1.3672	700	0.6895	-0.0025	-0.0108	0.4703	0.0083	-18.8379	-16.7630	-0.5991	-0.5989
0.6803	1.4648	750	0.6907	-0.0024	-0.0081	0.4242	0.0057	-18.8289	-16.7626	-0.5983	-0.5982
0.6836	1.5625	800	0.6911	-0.0028	-0.0078	0.4549	0.0050	-18.8281	-16.7640	-0.5989	-0.5987
0.6774	1.6602	850	0.6904	-0.0039	-0.0103	0.4484	0.0064	-18.8363	-16.7677	-0.5988	-0.5987
0.6866	1.7578	900	0.6875	-0.0009	-0.0130	0.4769	0.0121	-18.8454	-16.7576	-0.5987	-0.5986
0.6811	1.8555	950	0.6892	-0.0034	-0.0121	0.4725	0.0086	-18.8422	-16.7661	-0.5986	-0.5984
0.6812	1.9531	1000	0.6892	-0.0034	-0.0121	0.4725	0.0086	-18.8422	-16.7661	-0.5986	-0.5984

Framework versions

Transformers 4.40.1
Pytorch 2.0.0+cu117
Datasets 2.19.1
Tokenizers 0.19.1

tsavage68/chat_1000_STEPS_03beta_1e7rate_CDPOSFT

chat_1000_STEPS_03beta_1e7rate_CDPOSFT

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
chat_1000_STEPS_03beta_1e7rate_CDPOSFT