IE_L3_150steps_1e7rate_01beta_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: 0.2049
• Rewards/chosen: -0.3084
• Rewards/rejected: -2.8821
• Rewards/accuracies: 0.7400
• Rewards/margins: 2.5737
• Logps/rejected: -104.4487
• Logps/chosen: -85.8823
• Logits/rejected: -0.7993
• Logits/chosen: -0.7183

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-07
• 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: 150

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6292	0.4	50	0.5972	-0.0178	-0.2247	0.7400	0.2070	-77.8745	-82.9754	-0.7952	-0.7369
0.2432	0.8	100	0.2531	-0.1984	-1.9084	0.7400	1.7099	-94.7109	-84.7823	-0.7935	-0.7222
0.1679	1.2	150	0.2049	-0.3084	-2.8821	0.7400	2.5737	-104.4487	-85.8823	-0.7993	-0.7183

Framework versions

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