Llama-3.1-8B-Instruct-SAA-200

This model is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct on the bct_non_cot_dpo_200 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4427
• Rewards/chosen: -0.0414
• Rewards/rejected: -0.1125
• Rewards/accuracies: 0.9000
• Rewards/margins: 0.0711
• Logps/rejected: -1.1247
• Logps/chosen: -0.4137
• Logits/rejected: -0.4660
• Logits/chosen: -0.3670
• Sft Loss: 0.0454
• Odds Ratio Loss: 3.9738

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Sft Loss	Odds Ratio Loss
0.978	4.4444	50	0.8369	-0.0801	-0.1461	0.9500	0.0660	-1.4606	-0.8010	-0.4790	-0.3809	0.1018	7.3506
0.5508	8.8889	100	0.4427	-0.0414	-0.1125	0.9000	0.0711	-1.1247	-0.4137	-0.4660	-0.3670	0.0454	3.9738

Framework versions

• PEFT 0.12.0
• Transformers 4.45.2
• Pytorch 2.3.0
• Datasets 2.19.0
• Tokenizers 0.20.0