IE_L3_350steps_1e8rate_03beta_cSFTDPO

This model is a fine-tuned version of tsavage68/IE_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6896
• Rewards/chosen: -0.0071
• Rewards/rejected: -0.0198
• Rewards/accuracies: 0.4400
• Rewards/margins: 0.0127
• Logps/rejected: -75.6932
• Logps/chosen: -82.8214
• Logits/rejected: -0.7977
• Logits/chosen: -0.7408

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
0.6912	0.4	50	0.6940	-0.0075	-0.0104	0.4000	0.0029	-75.6618	-82.8226	-0.7964	-0.7393
0.6947	0.8	100	0.6925	0.0014	-0.0057	0.3850	0.0070	-75.6461	-82.7931	-0.7963	-0.7394
0.6881	1.2	150	0.7003	-0.0102	-0.0020	0.375	-0.0082	-75.6340	-82.8318	-0.7969	-0.7398
0.6776	1.6	200	0.6938	-0.0057	-0.0098	0.375	0.0041	-75.6601	-82.8168	-0.7970	-0.7399
0.6859	2.0	250	0.6850	-0.0033	-0.0250	0.4350	0.0217	-75.7105	-82.8087	-0.7975	-0.7405
0.7024	2.4	300	0.6893	-0.0075	-0.0207	0.4400	0.0132	-75.6964	-82.8228	-0.7977	-0.7408
0.6802	2.8	350	0.6896	-0.0071	-0.0198	0.4400	0.0127	-75.6932	-82.8214	-0.7977	-0.7408

Framework versions

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