htlou/mm-interp-AA_preference_cocour_new_step10_0_60

AA_preference_cocour_new_step10_0_60

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_cocour_new_step10_0_60 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5132
• Rewards/chosen: 1.2483
• Rewards/rejected: -1.4384
• Rewards/accuracies: 0.8090
• Rewards/margins: 2.6868
• Logps/rejected: -217.5475
• Logps/chosen: -242.1045
• Logits/rejected: -2.5237
• Logits/chosen: -2.5348

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
0.5367	0.6231	50	0.5755	1.1959	-0.1615	0.7604	1.3575	-204.7787	-242.6288	-2.3101	-2.3204
0.2278	1.2461	100	0.5325	1.6096	-0.6681	0.7986	2.2777	-209.8443	-238.4921	-2.6060	-2.6072
0.2926	1.8692	150	0.5151	1.0491	-1.4143	0.8194	2.4633	-217.3059	-244.0971	-2.4769	-2.4878
0.1423	2.4922	200	0.5126	1.3148	-1.3120	0.8125	2.6268	-216.2832	-241.4400	-2.5411	-2.5506

Framework versions

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