Llama-3.1-8B-Instruct-SAA-300

This model is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct on the bct_non_cot_dpo_300 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2542
• Rewards/chosen: -0.0199
• Rewards/rejected: -0.0490
• Rewards/accuracies: 0.7333
• Rewards/margins: 0.0291
• Logps/rejected: -0.4904
• Logps/chosen: -0.1993
• Logits/rejected: -0.4804
• Logits/chosen: -0.4435
• Sft Loss: 0.0199
• Odds Ratio Loss: 2.3427

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
1.0613	2.9630	50	0.9214	-0.0866	-0.1163	0.7667	0.0297	-1.1632	-0.8661	-0.4969	-0.4510	0.0877	8.3368
0.3375	5.9259	100	0.3373	-0.0282	-0.0580	0.7333	0.0297	-0.5795	-0.2823	-0.4847	-0.4453	0.0271	3.1020
0.2209	8.8889	150	0.2542	-0.0199	-0.0490	0.7333	0.0291	-0.4904	-0.1993	-0.4804	-0.4435	0.0199	2.3427

Framework versions

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