htlou/backup_0202_llamafactory_AA_preference_Cherry_0_40-llava-mistral

AA_preference_Cherry_0_40

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_Cherry_0_40 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5449
• Rewards/chosen: 2.9584
• Rewards/rejected: -0.1862
• Rewards/accuracies: 0.8300
• Rewards/margins: 3.1446
• Logps/rejected: -260.6278
• Logps/chosen: -298.7855
• Logits/rejected: -1.9543
• Logits/chosen: -1.9311

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
0.5108	0.7373	40	0.4996	2.8014	0.7843	0.8000	2.0171	-250.9224	-300.3552	-1.8153	-1.8277
0.1548	1.4747	80	0.5516	2.5892	-0.3111	0.8150	2.9003	-261.8770	-302.4771	-1.7896	-1.7813
0.0739	2.2120	120	0.5419	2.8910	-0.1778	0.8150	3.0689	-260.5441	-299.4589	-2.0237	-1.9931
0.0785	2.9493	160	0.5450	2.9577	-0.1873	0.8250	3.1451	-260.6390	-298.7917	-1.9546	-1.9315

Framework versions

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