Meta-Llama-3-70B-Instruct

This model is a fine-tuned version of meta-llama/Meta-Llama-3-70B-Instruct on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 1.2884
• Rewards/chosen: -0.0888
• Rewards/rejected: -0.1138
• Rewards/accuracies: 0.6132
• Rewards/margins: 0.0250
• Logps/rejected: -1.1382
• Logps/chosen: -0.8884
• Logits/rejected: -0.0033
• Logits/chosen: 0.2012
• Nll Loss: 1.2075
• Log Odds Ratio: -0.6278
• Log Odds Chosen: 0.3768

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: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

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
1.2483	0.9999	3555	1.2884	-0.0888	-0.1138	0.6132	0.0250	-1.1382	-0.8884	-0.0033	0.2012	1.2075	-0.6278	0.3768

Framework versions

• PEFT 0.11.1
• Transformers 4.41.0
• Pytorch 2.2.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1