htlou/backup_0202_llamafactory_AA_preference_Cherry_0_90-llava-mistral

AA_preference_Cherry_0_90

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_Cherry_0_90 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5233
• Rewards/chosen: 0.7968
• Rewards/rejected: -1.8577
• Rewards/accuracies: 0.8218
• Rewards/margins: 2.6545
• Logps/rejected: -234.1042
• Logps/chosen: -250.9109
• Logits/rejected: -1.8912
• Logits/chosen: -1.9506

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: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• 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
0.4674	0.8230	100	0.4985	0.1779	-1.6218	0.7940	1.7997	-231.7454	-257.0998	-2.2566	-2.2803
0.2463	1.6461	200	0.5190	1.1459	-1.2068	0.7963	2.3527	-227.5959	-247.4204	-1.6292	-1.7063
0.1311	2.4691	300	0.5212	0.6944	-1.9013	0.8194	2.5958	-234.5411	-251.9348	-1.8631	-1.9243

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.20.3