htlou/mm-interp-AA_preference_random_0_60

AA_preference_random_0_60

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_random_0_60 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5970
• Rewards/chosen: 1.1265
• Rewards/rejected: -0.9790
• Rewards/accuracies: 0.7882
• Rewards/margins: 2.1055
• Logps/rejected: -220.6737
• Logps/chosen: -235.6061
• Logits/rejected: -2.2225
• Logits/chosen: -2.2440

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
0.5547	0.6231	50	0.5781	0.9538	-0.0749	0.7222	1.0286	-211.6319	-237.3329	-2.4876	-2.4892
0.2103	1.2461	100	0.6054	1.3022	-0.3360	0.7778	1.6381	-214.2431	-233.8492	-2.2990	-2.3139
0.2095	1.8692	150	0.5998	1.4071	-0.5239	0.7743	1.9310	-216.1227	-232.8000	-2.3737	-2.3872
0.1498	2.4922	200	0.5972	1.0916	-1.0086	0.7847	2.1001	-220.9690	-235.9549	-2.2258	-2.2469

Framework versions

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