Llama-3-8B-Taiwan-Llawa-TCxYZL-DPO-Beta-0.01-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: 0.5434
• Rewards/chosen: -0.3119
• Rewards/rejected: -0.7953
• Rewards/accuracies: 0.8245
• Rewards/margins: 0.4833
• Logps/rejected: -118.7237
• Logps/chosen: -150.8103
• Logits/rejected: -0.1068
• Logits/chosen: -0.0702

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-07
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 6
• gradient_accumulation_steps: 64
• total_train_batch_size: 384
• total_eval_batch_size: 6
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• 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
No log	0	0	0.6931	0.0	0.0	0.0	0.0	-39.1966	-119.6160	-0.1592	-0.1378
0.6468	0.9937	76	0.6482	-0.0350	-0.1333	0.8596	0.0983	-52.5218	-123.1127	-0.2376	-0.2140
0.5604	1.9873	152	0.5646	-0.2442	-0.6292	0.8245	0.3850	-102.1200	-144.0388	-0.1309	-0.0973
0.5142	2.9810	228	0.5434	-0.3119	-0.7953	0.8245	0.4833	-118.7237	-150.8103	-0.1068	-0.0702

Framework versions

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