htlou/mm-interp-AA_preference_l0_0_50

AA_preference_l0_0_50

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_l0_0_50 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5682
• Rewards/chosen: 1.6580
• Rewards/rejected: -0.6999
• Rewards/accuracies: 0.7958
• Rewards/margins: 2.3579
• Logps/rejected: -234.7341
• Logps/chosen: -259.9489
• Logits/rejected: -2.1925
• Logits/chosen: -2.2317

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
0.5987	0.7463	50	0.6138	1.2435	-0.1587	0.7583	1.4022	-229.3223	-264.0939	-2.0743	-2.1097
0.2776	1.4925	100	0.5807	1.7478	-0.3381	0.75	2.0860	-231.1164	-259.0504	-2.1973	-2.2346
0.1711	2.2388	150	0.5595	1.4504	-0.7487	0.7833	2.1991	-235.2227	-262.0249	-2.1687	-2.2093
0.1884	2.9851	200	0.5684	1.6586	-0.7002	0.7917	2.3588	-234.7375	-259.9425	-2.1927	-2.2319

Framework versions

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