RAY2L/Llama-3-Instruct-8B-SimPOW-1

llama-3-8b-instruct-simpo

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the princeton-nlp/llama3-ultrafeedback dataset. It achieves the following results on the evaluation set:

• Loss: 0.7528
• Original Losses: 2.0491
• Weight: 0.3713
• Abs Diff: 3.1759
• Rewards/chosen: -45.3959
• Rewards/rejected: -50.3664
• Rewards/accuracies: 0.6976
• Rewards/margins: 4.9705
• Logps/rejected: -20.1465
• Logps/chosen: -18.1584
• Logits/rejected: 1.8309
• Logits/chosen: 1.7177
• All Logps 1: -7614.6904
• All Logps 1 Values: -7614.6909
• All Logps 2: 414.8609
• All Logps 2 Values: 414.8609

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: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_batch_size: 32
• 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	Original Losses	Weight	Abs Diff	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	All Logps 1	All Logps 1 Values	All Logps 2	All Logps 2 Values
0.7506	0.8549	400	0.7528	2.0491	0.3713	3.1759	-45.3959	-50.3664	0.6976	4.9705	-20.1465	-18.1584	1.8309	1.7177	-7614.6904	-7614.6909	414.8609	414.8609

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.2+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1