tsavage68/mpt_1000_STEPS_1e8_rate03_beta_DPO

mpt_1000_STEPS_1e5_rate_03_beta_DPO

This model is a fine-tuned version of mosaicml/mpt-7b-instruct on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6933
• Rewards/chosen: -0.0008
• Rewards/rejected: -0.0019
• Rewards/accuracies: 0.5187
• Rewards/margins: 0.0011
• Logps/rejected: -21.5638
• Logps/chosen: -20.7947
• Logits/rejected: 14.2524
• Logits/chosen: 14.2550

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-08
• train_batch_size: 2
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• 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
0.6965	0.1	100	0.6951	-0.0017	0.0013	0.4681	-0.0029	-21.5532	-20.7977	14.2557	14.2583
0.6918	0.2	200	0.6942	-0.0054	-0.0044	0.5011	-0.0010	-21.5722	-20.8104	14.2575	14.2601
0.6965	0.29	300	0.6941	-0.0016	-0.0010	0.4945	-0.0006	-21.5608	-20.7975	14.2549	14.2575
0.6906	0.39	400	0.6946	0.0001	0.0020	0.4747	-0.0019	-21.5507	-20.7919	14.2494	14.2520
0.6883	0.49	500	0.6972	-0.0019	0.0050	0.4484	-0.0069	-21.5408	-20.7986	14.2521	14.2547
0.6867	0.59	600	0.6969	-0.0054	0.0010	0.4418	-0.0064	-21.5541	-20.8103	14.2502	14.2528
0.6937	0.68	700	0.6939	0.0015	0.0020	0.5275	-0.0005	-21.5508	-20.7871	14.2547	14.2573
0.6855	0.78	800	0.6933	-0.0008	-0.0017	0.5099	0.0009	-21.5631	-20.7947	14.2522	14.2548
0.6918	0.88	900	0.6933	-0.0008	-0.0019	0.5187	0.0011	-21.5638	-20.7947	14.2524	14.2550
0.6957	0.98	1000	0.6933	-0.0008	-0.0019	0.5187	0.0011	-21.5638	-20.7947	14.2524	14.2550

Framework versions

• Transformers 4.39.1
• Pytorch 2.0.0+cu117
• Datasets 2.18.0
• Tokenizers 0.15.2