Llama0-3-8b-v0.1-p-0.05-lr6e-7-e1

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: 0.6348
• Rewards/chosen: -0.7856
• Rewards/rejected: -0.8817
• Rewards/accuracies: 0.5766
• Rewards/margins: 0.0961
• Logps/rejected: -174.9295
• Logps/chosen: -166.8561
• Logits/rejected: 0.2248
• Logits/chosen: 0.2123

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: 6e-07
• 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: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6639	0.2137	100	0.6632	-0.0650	-0.0751	0.5968	0.0101	-94.2649	-94.7914	0.0767	0.0572
0.6507	0.4275	200	0.6492	-0.2974	-0.3355	0.6008	0.0381	-120.3051	-118.0318	0.1386	0.1215
0.6383	0.6412	300	0.6397	-0.6120	-0.6852	0.5887	0.0732	-155.2713	-149.4875	0.2203	0.2063
0.6362	0.8549	400	0.6356	-0.7457	-0.8367	0.5766	0.0910	-170.4306	-162.8660	0.2216	0.2081

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.20.0