This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct on the princeton-nlp/llama3-ultrafeedback dataset. It achieves the following results on the evaluation set:

Loss: 1.5984
Rewards/chosen: -14.3945
Rewards/rejected: -15.5836
Rewards/accuracies: 0.6304
Rewards/margins: 1.1892
Logps/rejected: -155.8365
Logps/chosen: -143.9448
Logits/rejected: -0.3142
Logits/chosen: -0.3408
Nll Loss: 0.3937

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: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 8
total_train_batch_size: 128
total_eval_batch_size: 16
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
num_epochs: 3

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
1.5867	0.9986	432	1.5248	-16.0094	-16.9746	0.6587	0.9652	-169.7457	-160.0941	-0.4783	-0.5128	0.4373
0.7108	1.9994	865	1.5252	-14.8375	-15.9459	0.6500	1.1084	-159.4588	-148.3749	-0.4403	-0.4684	0.4056
0.4426	2.9957	1296	1.5984	-14.3945	-15.5836	0.6304	1.1892	-155.8365	-143.9448	-0.3142	-0.3408	0.3937

Framework versions

Transformers 4.44.2
Pytorch 2.3.1
Datasets 2.21.0
Tokenizers 0.19.1

RichardErkhov/jbjeong91_-_llama3.1-cpo-full-0911-4bits

Original model description:

llama3.1-cpo-full-0911

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

RichardErkhov/
jbjeong91_-_llama3.1-cpo-full-0911-4bits