smb-vision-base-1029

This model is trained from scratch using VideoMAE on over 4.7k CT volumes.

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: 3e-04
• train_batch_size: 32
• eval_batch_size: 1
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• num_epochs: 30.0

Training results

{ "_runtime": 54805.860011105, "_step": 4351, "eval/runtime": 17.8428, "eval/samples_per_second": 2.578, "eval/steps_per_second": 2.578, "total_flos": 3.8084565648770335e+21, "train/epoch": 30, "train/global_step": 4350, "train/grad_norm": 0.0735374316573143, "train/learning_rate": 0, "train/loss": 0.5736, "train_loss": 0.5022664608695041, "train_runtime": 54785.1298, "train_samples_per_second": 2.527, "train_steps_per_second": 0.079 }

Framework versions

• Transformers 4.46.0
• Pytorch 2.5.0
• Datasets 3.0.2
• Tokenizers 0.20.1

How to use

# load data using `dataload.py`

model = VideoMAEForPreTraining.from_pretrained(
    standardmodelbio/smb-vision-base,
    trust_remote_code=True,
)

embedding = model.videomae(batch["image"])