mohsinmirzax/smolvla_pick_place_sharpner_model

SmolVLA Pick & Place Sharpener

A fine-tuned vision-language action (VLA) model trained on pick-and-place tasks using the LeRobot framework.

Model Details

• Framework: LeRobot
• Task: Pick & Place (Wrist-only)
• Training Steps: 15,000
• Model Format: SafeTensors
• Base Model: SmolVLA

Files

Model Weights

• pretrained_model/model.safetensors - Main model weights
• pretrained_model/config.json - Model architecture configuration
• pretrained_model/train_config.json - Training hyperparameters

Preprocessing & Postprocessing

• pretrained_model/policy_preprocessor.json - Input normalization config
• pretrained_model/policy_preprocessor_step_5_normalizer_processor.safetensors - Normalizer weights
• pretrained_model/policy_postprocessor.json - Output denormalization config
• pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors - Denormalizer weights

Training State

• training_state/ - Optimizer and scheduler states for resuming training

Training Data

This model was trained on the dataset available at: mohsinmirzax/smolvla_pick_place_sharpner

Usage

from lerobot.common.policies.diffusion_policy import DiffusionPolicy

# Load the model
policy = DiffusionPolicy.from_pretrained(
    "mohsinmirzax/smolvla_pick_place_sharpner_model",
    subfolder="pretrained_model"
)

# Use for inference
policy.eval()
output = policy(observations)  # observations should be preprocessed

Training Logs

WandB training metrics and logs are available in the wandb_logs/ directory.

License

MIT

Citation

If you use this model, please cite the LeRobot framework and the original training dataset.