llama3_false_positives_1101_KTO_optimised_model

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5288
• Rewards/chosen: 0.5981
• Logps/chosen: -45.9608
• Rewards/rejected: -0.3258
• Logps/rejected: -56.8765
• Rewards/margins: 0.9238
• Kl: 0.0439

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-05
• train_batch_size: 1
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 6.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Logps/chosen	Rewards/rejected	Logps/rejected	Rewards/margins	Kl
0.4886	0.96	12	0.6709	0.1420	-50.5211	0.0978	-52.6405	0.0442	0.1381
0.5124	2.0	25	0.5966	0.3060	-48.8816	-0.1760	-55.3786	0.4819	0.0375
0.3647	2.96	37	0.5552	0.4908	-47.0331	-0.2555	-56.1745	0.7464	0.0447
0.3725	4.0	50	0.5239	0.5506	-46.4352	-0.3964	-57.5829	0.9470	0.0317
0.3249	4.96	62	0.5300	0.5839	-46.1019	-0.3309	-56.9280	0.9148	0.0409
0.3262	5.76	72	0.5288	0.5981	-45.9608	-0.3258	-56.8765	0.9238	0.0439

Framework versions

• PEFT 0.11.1
• Transformers 4.37.2
• Pytorch 2.2.0
• Datasets 3.1.0
• Tokenizers 0.15.2