Llama-3.1-8B-Instruct-SAA-800

This model is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct on the bct_non_cot_dpo_800 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1529
• Rewards/chosen: -0.0121
• Rewards/rejected: -0.0692
• Rewards/accuracies: 0.8125
• Rewards/margins: 0.0571
• Logps/rejected: -0.6922
• Logps/chosen: -0.1209
• Logits/rejected: -0.3690
• Logits/chosen: -0.3184
• Sft Loss: 0.0162
• Odds Ratio Loss: 1.3676

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
1.4945	1.1111	50	1.2763	-0.1228	-0.1709	0.7875	0.0481	-1.7095	-1.2281	-0.4548	-0.3794	0.1502	11.2609
0.2666	2.2222	100	0.2491	-0.0212	-0.0729	0.8250	0.0517	-0.7288	-0.2119	-0.4319	-0.3666	0.0242	2.2484
0.1014	3.3333	150	0.1632	-0.0129	-0.0606	0.8125	0.0477	-0.6058	-0.1292	-0.3820	-0.3284	0.0168	1.4635
0.1429	4.4444	200	0.1534	-0.0121	-0.0584	0.8125	0.0463	-0.5841	-0.1211	-0.3818	-0.3298	0.0158	1.3752
0.1007	5.5556	250	0.1530	-0.0121	-0.0641	0.8125	0.0520	-0.6407	-0.1206	-0.3740	-0.3235	0.0159	1.3704
0.1385	6.6667	300	0.1534	-0.0122	-0.0688	0.8000	0.0566	-0.6881	-0.1217	-0.3725	-0.3214	0.0161	1.3729
0.0918	7.7778	350	0.1537	-0.0122	-0.0689	0.8125	0.0567	-0.6889	-0.1217	-0.3698	-0.3191	0.0162	1.3742
0.0752	8.8889	400	0.1534	-0.0121	-0.0690	0.8000	0.0568	-0.6896	-0.1213	-0.3706	-0.3195	0.0162	1.3723
0.1052	10.0	450	0.1529	-0.0121	-0.0692	0.8125	0.0571	-0.6922	-0.1209	-0.3690	-0.3184	0.0162	1.3676

Framework versions

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