chat_400_STEPS_01beta_5e7rate_CDPOSFT

This model is a fine-tuned version of tsavage68/chat_600STEPS_1e8rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6851
• Rewards/chosen: -0.0303
• Rewards/rejected: -0.0485
• Rewards/accuracies: 0.5077
• Rewards/margins: 0.0182
• Logps/rejected: -19.2868
• Logps/chosen: -17.0576
• Logits/rejected: -0.6041
• Logits/chosen: -0.6040

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: 5e-07
• train_batch_size: 4
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 400

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6924	0.0977	50	0.6933	0.0017	0.0020	0.4154	-0.0003	-18.7815	-16.7372	-0.5990	-0.5988
0.6889	0.1953	100	0.6896	-0.0103	-0.0178	0.4769	0.0075	-18.9805	-16.8580	-0.6027	-0.6025
0.692	0.2930	150	0.6885	-0.0339	-0.0443	0.4967	0.0104	-19.2452	-17.0936	-0.6039	-0.6038
0.6898	0.3906	200	0.6871	-0.0252	-0.0389	0.5033	0.0137	-19.1906	-17.0066	-0.6024	-0.6022
0.6911	0.4883	250	0.6862	-0.0287	-0.0445	0.5099	0.0159	-19.2474	-17.0415	-0.6037	-0.6036
0.6854	0.5859	300	0.6852	-0.0303	-0.0482	0.5121	0.0179	-19.2838	-17.0573	-0.6047	-0.6046
0.683	0.6836	350	0.6849	-0.0303	-0.0489	0.5231	0.0186	-19.2907	-17.0575	-0.6039	-0.6037
0.6853	0.7812	400	0.6851	-0.0303	-0.0485	0.5077	0.0182	-19.2868	-17.0576	-0.6041	-0.6040

Framework versions

• Transformers 4.40.1
• Pytorch 2.0.0+cu117
• Datasets 2.19.1
• Tokenizers 0.19.1