Summary_L3_150steps_1e6rate_01beta_CSFTDPO

This model is a fine-tuned version of tsavage68/Summary_L3_1000steps_1e7rate_SFT2 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5961
• Rewards/chosen: -0.0077
• Rewards/rejected: -1.5938
• Rewards/accuracies: 0.1400
• Rewards/margins: 1.5861
• Logps/rejected: -31.2014
• Logps/chosen: -9.4593
• Logits/rejected: -1.2000
• Logits/chosen: -1.1980

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: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 4
• 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
0.571	0.2004	50	0.5986	0.0271	-0.6059	0.1400	0.6329	-21.3224	-9.1122	-1.1153	-1.1163
0.6585	0.4008	100	0.5962	0.0177	-1.2883	0.1400	1.3060	-28.1472	-9.2058	-1.1739	-1.1725
0.6238	0.6012	150	0.5961	-0.0077	-1.5938	0.1400	1.5861	-31.2014	-9.4593	-1.2000	-1.1980

Framework versions

• Transformers 4.41.2
• Pytorch 2.0.0+cu117
• Datasets 2.20.0
• Tokenizers 0.19.1