htlou/mm-interp-AA_preference_cocour_new_step10_0_20

AA_preference_cocour_new_step10_0_20

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_cocour_new_step10_0_20 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5375
• Rewards/chosen: 1.9424
• Rewards/rejected: -0.2063
• Rewards/accuracies: 0.8333
• Rewards/margins: 2.1487
• Logps/rejected: -192.4608
• Logps/chosen: -228.8406
• Logits/rejected: -2.4079
• Logits/chosen: -2.4243

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
0.2912	1.8692	50	0.5429	1.8078	-0.1404	0.7812	1.9482	-191.8026	-230.1866	-2.4138	-2.4286

Framework versions

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