htlou/mm-interp-AA_preference_l0_new_step10_0_60

AA_preference_l0_new_step10_0_60

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_l0_new_step10_0_60 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5930
• Rewards/chosen: 0.5295
• Rewards/rejected: -1.8896
• Rewards/accuracies: 0.7917
• Rewards/margins: 2.4190
• Logps/rejected: -242.2141
• Logps/chosen: -237.5954
• Logits/rejected: -2.0873
• Logits/chosen: -2.1246

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
0.5728	0.6231	50	0.5968	0.7303	-0.4694	0.7257	1.1997	-228.0123	-235.5866	-2.3844	-2.3915
0.2338	1.2461	100	0.6138	0.8569	-1.0717	0.7708	1.9286	-234.0351	-234.3208	-2.3760	-2.3908
0.2653	1.8692	150	0.5830	0.4703	-1.7333	0.7847	2.2035	-240.6510	-238.1874	-2.0378	-2.0764
0.1596	2.4922	200	0.5909	0.5892	-1.8017	0.7951	2.3909	-241.3353	-236.9982	-2.0980	-2.1340

Framework versions

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