Llama0-3-8b-ultra-p-0.05-lr1e-6-e3

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5678
• Rewards/chosen: -3.1024
• Rewards/rejected: -5.3878
• Rewards/accuracies: 0.7656
• Rewards/margins: 2.2855
• Logps/rejected: -803.3770
• Logps/chosen: -566.8709
• Logits/rejected: -0.7286
• Logits/chosen: -0.3055

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: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5867	0.2060	100	0.5790	-0.4168	-0.7897	0.6797	0.3729	-343.5607	-298.3142	0.0683	0.0050
0.5459	0.4119	200	0.5337	-0.8336	-1.5839	0.7422	0.7504	-422.9892	-339.9911	0.4007	0.2879
0.5201	0.6179	300	0.5116	-0.7067	-1.5136	0.7344	0.8069	-415.9542	-327.3016	0.3623	0.2661
0.5068	0.8239	400	0.5037	-0.7404	-1.6591	0.7891	0.9187	-430.5064	-330.6776	0.2848	0.2141
0.427	1.0299	500	0.5057	-1.4842	-2.9740	0.75	1.4898	-561.9933	-405.0575	-0.1430	-0.0848
0.3367	1.2358	600	0.5150	-1.9307	-3.6670	0.75	1.7363	-631.2911	-449.7062	-0.1170	-0.0016
0.336	1.4418	700	0.5013	-1.6315	-3.1525	0.7656	1.5211	-579.8499	-419.7817	-0.0661	0.0619
0.3443	1.6478	800	0.4919	-1.5274	-2.9336	0.7656	1.4062	-557.9580	-409.3778	-0.0808	0.0430
0.3387	1.8538	900	0.5136	-1.8875	-3.4761	0.7578	1.5886	-612.2042	-445.3885	-0.0675	0.0881
0.2045	2.0597	1000	0.5396	-2.6871	-4.6850	0.7656	1.9979	-733.0979	-525.3492	-0.3513	-0.1306
0.1911	2.2657	1100	0.5562	-3.0265	-5.1837	0.7422	2.1572	-782.9683	-559.2891	-0.6321	-0.2757
0.1935	2.4717	1200	0.5518	-2.8870	-5.0043	0.75	2.1173	-765.0246	-545.3388	-0.6105	-0.2462
0.1909	2.6777	1300	0.5623	-3.0447	-5.2451	0.75	2.2004	-789.1040	-561.1038	-0.6371	-0.2728
0.1805	2.8836	1400	0.5746	-3.2314	-5.5860	0.75	2.3546	-823.1945	-579.7725	-0.7721	-0.3436

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1