htlou/backup_0202_llamafactory_AA_preference_Cherry_0_20-llava-mistral

AA_preference_Cherry_0_20

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_Cherry_0_20 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4124
• Rewards/chosen: 3.4337
• Rewards/rejected: 0.0778
• Rewards/accuracies: 0.8558
• Rewards/margins: 3.3559
• Logps/rejected: -252.4232
• Logps/chosen: -315.7890
• Logits/rejected: -2.1962
• Logits/chosen: -2.2083

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
0.1046	1.4679	40	0.4104	3.1981	0.3078	0.8558	2.8903	-250.1229	-318.1455	-2.1406	-2.1573
0.0449	2.9358	80	0.4136	3.4294	0.0760	0.8558	3.3534	-252.4416	-315.8324	-2.1969	-2.2089

Framework versions

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