htlou/mm-interp-AA_preference_cocour_new_step10_0_50

AA_preference_cocour_new_step10_0_50

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_cocour_new_step10_0_50 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5010
• Rewards/chosen: 1.1422
• Rewards/rejected: -1.7885
• Rewards/accuracies: 0.8583
• Rewards/margins: 2.9307
• Logps/rejected: -214.9074
• Logps/chosen: -264.3012
• Logits/rejected: -2.0474
• Logits/chosen: -2.1072

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
0.6039	0.7463	50	0.5475	1.3230	-0.4024	0.8167	1.7254	-201.0464	-262.4934	-2.4300	-2.4488
0.2811	1.4925	100	0.5121	1.3770	-1.3342	0.8542	2.7112	-210.3646	-261.9533	-1.9351	-2.0044
0.1518	2.2388	150	0.5044	1.4880	-1.2566	0.8542	2.7446	-209.5883	-260.8434	-2.0397	-2.0982
0.1381	2.9851	200	0.5010	1.1421	-1.7916	0.8583	2.9338	-214.9391	-264.3022	-2.0472	-2.1071

Framework versions

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