htlou/backup_0202_llamafactory_AA_preference_cosi_new_step10_0_90-llava-mistral

AA_preference_cosi_new_step10_0_90

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_cosi_new_step10_0_90 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4606
• Rewards/chosen: 0.1088
• Rewards/rejected: -2.5083
• Rewards/accuracies: 0.7963
• Rewards/margins: 2.6171
• Logps/rejected: -247.9763
• Logps/chosen: -252.3918
• Logits/rejected: -2.0273
• Logits/chosen: -2.0758

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
0.5991	0.4158	50	0.5701	1.0641	0.0162	0.7199	1.0479	-222.7316	-242.8391	-2.2582	-2.2788
0.4822	0.8316	100	0.4792	0.2741	-1.4181	0.7708	1.6922	-237.0740	-250.7390	-2.2864	-2.3142
0.2575	1.2474	150	0.4869	0.5197	-1.5659	0.7731	2.0855	-238.5522	-248.2833	-2.1810	-2.2088
0.2918	1.6632	200	0.4748	0.6541	-1.5052	0.7894	2.1593	-237.9451	-246.9389	-2.1041	-2.1397
0.1553	2.0790	250	0.4500	0.2460	-2.1200	0.7917	2.3660	-244.0935	-251.0200	-2.0244	-2.0690
0.1681	2.4948	300	0.4637	0.1908	-2.3560	0.7917	2.5468	-246.4534	-251.5723	-2.0304	-2.0780
0.2027	2.9106	350	0.4606	0.1105	-2.5095	0.7940	2.6200	-247.9879	-252.3748	-2.0270	-2.0755

Framework versions

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