htlou/backup_0202_llamafactory_AA_preference_Cherry_0_30-llava-mistral

AA_preference_Cherry_0_30

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_Cherry_0_30 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5028
• Rewards/chosen: 3.6010
• Rewards/rejected: 0.2000
• Rewards/accuracies: 0.875
• Rewards/margins: 3.4010
• Logps/rejected: -228.0410
• Logps/chosen: -307.6524
• Logits/rejected: -2.3391
• Logits/chosen: -2.3614

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
0.474	0.9816	40	0.4699	3.1334	0.8359	0.8553	2.2975	-221.6830	-312.3291	-2.1664	-2.2103
0.1662	1.9632	80	0.5018	3.4522	0.2267	0.8618	3.2255	-227.7745	-309.1402	-2.3660	-2.3873
0.0328	2.9448	120	0.5028	3.6010	0.2000	0.875	3.4010	-228.0410	-307.6524	-2.3391	-2.3614

Framework versions

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