llama3-L1-SFT-L2-KTO

This model is a fine-tuned version of EllieS/TempReason-L1-llama3 on the EllieS/Temp-L2-DPO dataset. It achieves the following results on the evaluation set:

• Loss: 0.2122
• Rewards/chosen: 0.3257
• Rewards/rejected: -9.5548
• Rewards/accuracies: 1.0
• Rewards/margins: 9.8805
• Logps/rejected: -1018.5145
• Logps/chosen: -12.0858
• Logits/rejected: 1.0988
• Logits/chosen: 0.1932

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-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• total_eval_batch_size: 2
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.2129	0.4994	1000	0.2124	0.3252	-9.3514	1.0	9.6766	-998.1762	-12.1315	1.1081	0.2036
0.2118	0.9989	2000	0.2122	0.3257	-9.5548	1.0	9.8805	-1018.5145	-12.0858	1.0988	0.1932

Framework versions

• PEFT 0.7.1
• Transformers 4.40.2
• Pytorch 2.1.2+cu121
• Datasets 2.18.0
• Tokenizers 0.19.1