Llama-3-8B-Taiwan-Llawa-TCxYZL-ORPO-Beta-0.1-Instruct

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

• Loss: 0.9671
• Rewards/chosen: -0.0787
• Rewards/rejected: -0.1126
• Rewards/accuracies: 0.6191
• Rewards/margins: 0.0339
• Logps/rejected: -1.1257
• Logps/chosen: -0.7870
• Logits/rejected: -0.2509
• Logits/chosen: -0.2044
• Nll Loss: 0.8891
• Log Odds Ratio: -0.6148
• Log Odds Chosen: 0.4628

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: 1
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 6
• gradient_accumulation_steps: 64
• total_train_batch_size: 384
• 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
• num_epochs: 5.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Nll Loss	Log Odds Ratio	Log Odds Chosen
No log	0	0	1.0532	-0.1124	-0.0806	0.1183	-0.0318	-0.8062	-1.1242	-0.7965	-0.8056	0.9263	-1.1339	-0.6334
1.027	0.9937	76	1.0078	-0.0900	-0.0732	0.2985	-0.0168	-0.7319	-0.9002	-0.4705	-0.4640	0.8982	-0.9524	-0.3610
0.9555	1.9873	152	0.9834	-0.0812	-0.0860	0.4560	0.0048	-0.8598	-0.8116	-0.3296	-0.3117	0.8906	-0.7776	0.0042
0.9188	2.9941	229	0.9713	-0.0782	-0.1002	0.5555	0.0219	-1.0017	-0.7822	-0.2840	-0.2493	0.8883	-0.6676	0.2898
0.9388	3.9877	305	0.9677	-0.0784	-0.1101	0.5971	0.0317	-1.1007	-0.7836	-0.2566	-0.2131	0.8887	-0.6256	0.4299
0.9289	4.9683	380	0.9671	-0.0787	-0.1126	0.6191	0.0339	-1.1257	-0.7870	-0.2509	-0.2044	0.8891	-0.6148	0.4628

Framework versions

• Transformers 4.43.1
• Pytorch 2.3.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1