htlou/mm-interp-AA_preference_cocour_new_step10_0_40

AA_preference_cocour_new_step10_0_40

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_cocour_new_step10_0_40 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5211
• Rewards/chosen: 1.5578
• Rewards/rejected: -1.0138
• Rewards/accuracies: 0.8229
• Rewards/margins: 2.5717
• Logps/rejected: -198.0200
• Logps/chosen: -264.0636
• Logits/rejected: -2.3020
• Logits/chosen: -2.3288

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
0.5069	0.9346	50	0.5445	1.1549	-0.3300	0.7708	1.4849	-191.1814	-268.0926	-2.3323	-2.3625
0.33	1.8692	100	0.5182	1.7644	-0.6533	0.8229	2.4177	-194.4146	-261.9976	-2.1958	-2.2272
0.1617	2.8037	150	0.5213	1.5548	-1.0154	0.8177	2.5702	-198.0357	-264.0938	-2.3015	-2.3283

Framework versions

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