Gaston_dolphin-2.9.1-yi-1.5-9b

This model is a fine-tuned version of cognitivecomputations/dolphin-2.9.1-yi-1.5-9b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4290
• Rewards/chosen: -0.0153
• Rewards/rejected: -0.2895
• Rewards/accuracies: 0.9985
• Rewards/margins: 0.2742
• Logps/rejected: -2.8952
• Logps/chosen: -0.1528
• Logits/rejected: -0.1534
• Logits/chosen: 0.0002
• Nll Loss: 0.4278
• Log Odds Ratio: -0.0124
• Log Odds Chosen: 4.8981

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: 8e-06
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Nll Loss	Log Odds Ratio	Log Odds Chosen
0.5193	0.1005	103	0.5159	-0.0187	-0.0825	0.9971	0.0638	-0.8248	-0.1866	0.1547	0.1467	0.5004	-0.1555	2.0327
0.4988	0.2009	206	0.4724	-0.0170	-0.1413	0.9985	0.1243	-1.4130	-0.1703	0.0154	-0.0134	0.4661	-0.0627	3.0432
0.4375	0.3014	309	0.4577	-0.0162	-0.1628	0.9985	0.1466	-1.6283	-0.1622	0.1372	0.1328	0.4530	-0.0467	3.3955
0.4738	0.4019	412	0.4463	-0.0160	-0.2198	0.9985	0.2038	-2.1980	-0.1596	-0.0220	0.0649	0.4438	-0.0250	4.0928
0.4893	0.5023	515	0.4406	-0.0159	-0.2499	0.9985	0.2341	-2.4993	-0.1585	-0.0720	0.0474	0.4388	-0.0185	4.4389
0.4565	0.6028	618	0.4357	-0.0157	-0.3289	0.9985	0.3133	-3.2895	-0.1566	-0.1392	0.0470	0.4347	-0.0093	5.2916
0.4069	0.7032	721	0.4324	-0.0154	-0.3096	0.9985	0.2942	-3.0962	-0.1544	-0.1833	-0.0044	0.4313	-0.0107	5.1028
0.4297	0.8037	824	0.4299	-0.0153	-0.2854	0.9985	0.2701	-2.8536	-0.1528	-0.1911	-0.0397	0.4286	-0.0129	4.8536
0.4437	0.9042	927	0.4290	-0.0153	-0.2895	0.9985	0.2742	-2.8952	-0.1528	-0.1534	0.0002	0.4278	-0.0124	4.8981

Framework versions

• PEFT 0.11.1
• Transformers 4.41.0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1