htlou/mm-interp-RLAIF-V-L0-q0_25_preference

RLAIF-V-L0-q0_25_preference

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the RLAIF-V-L0-q0_25_preference dataset. It achieves the following results on the evaluation set:

• Loss: 0.5525
• Rewards/chosen: -2.5933
• Rewards/rejected: -4.2536
• Rewards/accuracies: 0.7344
• Rewards/margins: 1.6603
• Logps/rejected: -203.2228
• Logps/chosen: -183.7414
• Logits/rejected: -2.2176
• Logits/chosen: -2.2545

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
0.5342	0.6944	50	0.5209	-0.2071	-0.9491	0.7148	0.7420	-170.1774	-159.8791	-2.4678	-2.4995
0.217	1.3889	100	0.5167	-1.1670	-2.3865	0.7070	1.2195	-184.5513	-169.4779	-2.4044	-2.4350
0.105	2.0833	150	0.5275	-2.3365	-3.9689	0.7383	1.6324	-200.3754	-181.1734	-2.1635	-2.2023
0.0911	2.7778	200	0.5503	-2.5532	-4.2135	0.7227	1.6603	-202.8217	-183.3405	-2.2203	-2.2570

Framework versions

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