Syed-Hasan-8503/Llama-3-8b-instruct-SimPO

outputs

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the Syed-Hasan-8503/orpo-40k-train-test dataset. It achieves the following results on the evaluation set:

• Loss: 0.9751
• Rewards/chosen: -3.4539
• Rewards/rejected: -5.6604
• Rewards/accuracies: 0.7613
• Rewards/margins: 2.2065
• Logps/rejected: -2.2642
• Logps/chosen: -1.3816
• Logits/rejected: -1.3683
• Logits/chosen: -1.2117

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: 2
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 3
• gradient_accumulation_steps: 2
• total_train_batch_size: 12
• total_eval_batch_size: 12
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 200

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
1.645	0.0140	50	1.2563	-2.7945	-3.7325	0.7027	0.9380	-1.4930	-1.1178	-1.3468	-1.1841
0.8722	0.0280	100	1.0619	-3.0769	-4.7343	0.7320	1.6574	-1.8937	-1.2308	-1.3817	-1.2196
1.0404	0.0419	150	0.9883	-3.4545	-5.6160	0.7545	2.1615	-2.2464	-1.3818	-1.3639	-1.2082
1.4672	0.0559	200	0.9751	-3.4539	-5.6604	0.7613	2.2065	-2.2642	-1.3816	-1.3683	-1.2117

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.0+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1