chat_1000_STEPS_05beta_1e7rate_CDPOSFT

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: 0.6899
• Rewards/chosen: -0.0048
• Rewards/rejected: -0.0138
• Rewards/accuracies: 0.4527
• Rewards/margins: 0.0090
• Logps/rejected: -18.8295
• Logps/chosen: -16.7641
• Logits/rejected: -0.5988
• Logits/chosen: -0.5987

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
0.6929	0.0977	50	0.6947	-0.0000	0.0016	0.4066	-0.0016	-18.7989	-16.7547	-0.5985	-0.5983
0.694	0.1953	100	0.6903	0.0030	-0.0047	0.4527	0.0076	-18.8113	-16.7487	-0.5976	-0.5975
0.6922	0.2930	150	0.6941	-0.0056	-0.0053	0.4044	-0.0003	-18.8127	-16.7659	-0.5978	-0.5977
0.7012	0.3906	200	0.6957	-0.0099	-0.0065	0.4132	-0.0034	-18.8151	-16.7744	-0.5982	-0.5980
0.6992	0.4883	250	0.6932	-0.0081	-0.0099	0.4484	0.0017	-18.8217	-16.7709	-0.5975	-0.5974
0.6872	0.5859	300	0.6918	-0.0096	-0.0144	0.4440	0.0048	-18.8309	-16.7738	-0.5990	-0.5989
0.6875	0.6836	350	0.6894	-0.0116	-0.0209	0.4484	0.0093	-18.8438	-16.7778	-0.5985	-0.5984
0.6918	0.7812	400	0.6878	-0.0070	-0.0200	0.4462	0.0129	-18.8419	-16.7687	-0.5987	-0.5985
0.6868	0.8789	450	0.6897	-0.0052	-0.0141	0.4396	0.0089	-18.8302	-16.7651	-0.5982	-0.5981
0.6867	0.9766	500	0.6904	-0.0080	-0.0160	0.4176	0.0080	-18.8339	-16.7706	-0.5988	-0.5987
0.6744	1.0742	550	0.6883	-0.0035	-0.0157	0.4527	0.0123	-18.8334	-16.7616	-0.5985	-0.5984
0.6791	1.1719	600	0.6897	-0.0033	-0.0127	0.4484	0.0094	-18.8275	-16.7612	-0.5988	-0.5987
0.6793	1.2695	650	0.6887	-0.0077	-0.0191	0.4418	0.0114	-18.8402	-16.7700	-0.5985	-0.5983
0.6696	1.3672	700	0.6863	-0.0015	-0.0176	0.4527	0.0161	-18.8372	-16.7576	-0.5988	-0.5986
0.6689	1.4648	750	0.6873	-0.0024	-0.0167	0.4593	0.0143	-18.8353	-16.7594	-0.5983	-0.5982
0.6808	1.5625	800	0.6879	-0.0050	-0.0179	0.4637	0.0129	-18.8378	-16.7646	-0.5992	-0.5991
0.6718	1.6602	850	0.6902	-0.0058	-0.0139	0.4462	0.0082	-18.8299	-16.7662	-0.5985	-0.5984
0.678	1.7578	900	0.6872	-0.0008	-0.0151	0.4571	0.0144	-18.8323	-16.7562	-0.5989	-0.5988
0.6745	1.8555	950	0.6899	-0.0048	-0.0138	0.4527	0.0090	-18.8295	-16.7641	-0.5988	-0.5987
0.6759	1.9531	1000	0.6899	-0.0048	-0.0138	0.4527	0.0090	-18.8295	-16.7641	-0.5988	-0.5987

Framework versions

• Transformers 4.40.1
• Pytorch 2.0.0+cu117
• Datasets 2.19.1
• Tokenizers 0.19.1