htlou/mm-interp-RLAIF-V-Coccur-q0_25_preference

RLAIF-V-Coccur-q0_25_preference

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the RLAIF-V-Coccur-q0_25_preference dataset. It achieves the following results on the evaluation set:

• Loss: 0.5440
• Rewards/chosen: -2.3911
• Rewards/rejected: -4.1460
• Rewards/accuracies: 0.7188
• Rewards/margins: 1.7550
• Logps/rejected: -201.5194
• Logps/chosen: -183.7933
• Logits/rejected: -2.7049
• Logits/chosen: -2.7311

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
0.5904	0.6944	50	0.5585	0.1767	-0.5308	0.6953	0.7075	-165.3675	-158.1156	-2.7400	-2.7505
0.2124	1.3889	100	0.5330	-0.9785	-2.2414	0.7344	1.2630	-182.4733	-169.6674	-2.6846	-2.7028
0.1027	2.0833	150	0.5209	-1.3289	-2.7382	0.7305	1.4093	-187.4415	-173.1719	-2.7648	-2.7841
0.0793	2.7778	200	0.5435	-2.3758	-4.1313	0.7227	1.7554	-201.3717	-183.6412	-2.7055	-2.7316

Framework versions

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