tsavage68/mpt_1000_STEPS_1e6_rate_05_beta_DPO

mpt_1000_STEPS_1e8_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: 0.6941
• Rewards/chosen: -1.2875
• Rewards/rejected: -1.6132
• Rewards/accuracies: 0.6154
• Rewards/margins: 0.3257
• Logps/rejected: -24.7839
• Logps/chosen: -23.3672
• Logits/rejected: 14.1648
• Logits/chosen: 14.1681

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: 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
0.7473	0.1	100	0.6927	0.1811	0.1159	0.5582	0.0651	-21.3256	-20.4301	14.3166	14.3195
0.7098	0.2	200	0.7624	0.6571	0.5345	0.5714	0.1226	-20.4884	-19.4780	14.1537	14.1566
0.7516	0.29	300	0.7505	-0.8487	-1.0927	0.5429	0.2440	-23.7428	-22.4895	14.5590	14.5620
0.7762	0.39	400	0.7476	-2.2343	-2.4798	0.5692	0.2455	-26.5171	-25.2608	14.0064	14.0094
0.8328	0.49	500	0.7228	-1.5283	-1.7877	0.5736	0.2594	-25.1329	-23.8488	14.1811	14.1843
0.625	0.59	600	0.7006	-1.3183	-1.6353	0.5978	0.3170	-24.8281	-23.4288	14.3453	14.3486
0.7164	0.68	700	0.7015	-1.2944	-1.6029	0.6022	0.3084	-24.7632	-23.3811	14.2239	14.2271
0.6844	0.78	800	0.6985	-1.2758	-1.5914	0.6198	0.3157	-24.7403	-23.3437	14.1630	14.1663
0.6996	0.88	900	0.6971	-1.2896	-1.6092	0.6110	0.3196	-24.7758	-23.3713	14.1673	14.1706
0.6352	0.98	1000	0.6941	-1.2875	-1.6132	0.6154	0.3257	-24.7839	-23.3672	14.1648	14.1681

Framework versions

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