htlou/mm-interp-AA_preference_random_0_50

AA_preference_random_0_50

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_random_0_50 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6024
• Rewards/chosen: 1.2815
• Rewards/rejected: -0.9739
• Rewards/accuracies: 0.7917
• Rewards/margins: 2.2554
• Logps/rejected: -218.7351
• Logps/chosen: -248.4081
• Logits/rejected: -2.2174
• Logits/chosen: -2.2436

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
0.5788	0.7463	50	0.5886	1.3112	-0.0126	0.7667	1.3238	-209.1226	-248.1118	-2.3816	-2.3948
0.2623	1.4925	100	0.6057	1.5827	-0.4999	0.8042	2.0826	-213.9953	-245.3966	-2.2436	-2.2727
0.1176	2.2388	150	0.5923	1.4688	-0.7137	0.7875	2.1825	-216.1334	-246.5355	-2.2235	-2.2485
0.1387	2.9851	200	0.6027	1.2815	-0.9739	0.7875	2.2554	-218.7351	-248.4080	-2.2176	-2.2437

Framework versions

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