IE_M2_350steps_1e8rate_03beta_cSFTDPO

This model is a fine-tuned version of tsavage68/IE_M2_1000steps_1e7rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6746
• Rewards/chosen: -0.0013
• Rewards/rejected: -0.0404
• Rewards/accuracies: 0.3600
• Rewards/margins: 0.0391
• Logps/rejected: -41.1564
• Logps/chosen: -42.2098
• Logits/rejected: -2.9159
• Logits/chosen: -2.8545

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: 350

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6998	0.4	50	0.6949	0.0058	0.0085	0.2050	-0.0028	-40.9934	-42.1863	-2.9160	-2.8547
0.6925	0.8	100	0.6906	0.0017	-0.0041	0.2600	0.0059	-41.0355	-42.1997	-2.9159	-2.8546
0.679	1.2	150	0.6779	0.0047	-0.0273	0.3650	0.0320	-41.1127	-42.1899	-2.9158	-2.8546
0.6715	1.6	200	0.6747	0.0020	-0.0367	0.3900	0.0387	-41.1442	-42.1988	-2.9156	-2.8544
0.6764	2.0	250	0.6736	-0.0012	-0.0419	0.3850	0.0407	-41.1614	-42.2094	-2.9156	-2.8543
0.6842	2.4	300	0.6763	-0.0024	-0.0380	0.3500	0.0355	-41.1483	-42.2137	-2.9159	-2.8545
0.6712	2.8	350	0.6746	-0.0013	-0.0404	0.3600	0.0391	-41.1564	-42.2098	-2.9159	-2.8545

Framework versions

• Transformers 4.44.2
• Pytorch 2.0.0+cu117
• Datasets 3.0.0
• Tokenizers 0.19.1