htlou/mm-interp-AA_preference_cosi_new_step10_0_60

AA_preference_cosi_new_step10_0_60

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_cosi_new_step10_0_60 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5341
• Rewards/chosen: 1.2512
• Rewards/rejected: -1.3631
• Rewards/accuracies: 0.7951
• Rewards/margins: 2.6143
• Logps/rejected: -235.6757
• Logps/chosen: -268.5645
• Logits/rejected: -2.3102
• Logits/chosen: -2.3523

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
0.5451	0.6231	50	0.5847	1.2566	0.0476	0.7465	1.2091	-221.5687	-268.5103	-2.4265	-2.4445
0.2411	1.2461	100	0.5331	1.0138	-1.1612	0.7882	2.1750	-233.6565	-270.9385	-2.5634	-2.5859
0.2838	1.8692	150	0.5178	1.2878	-1.1255	0.7986	2.4133	-233.2990	-268.1986	-2.3948	-2.4266
0.1415	2.4922	200	0.5325	1.3080	-1.2671	0.8090	2.5751	-234.7151	-267.9961	-2.3379	-2.3777

Framework versions

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