mlfoundations-cua-dev/uitars_500_steps_gbs_8_wd_0.1_max_grad_norm_1.0_baseline

idm_tars_1.5_7b_frame_pairs_896x896_lr_1e-5_10_epochs_500_steps_gbs_8_wd_0.1_max_grad_norm_1.0_baseline

Model Information

Full Model Name: idm_tars_1.5_7b_frame_pairs_896x896_lr_1e-5_10_epochs_500_steps_gbs_8_wd_0.1_max_grad_norm_1.0_baseline

Repository Name: mlfoundations-cua-dev/uitars_500_steps_gbs_8_wd_0.1_max_grad_norm_1.0_baseline

Model Directory: idm_tars_1.5_7b_frame_pairs_896x896_lr_1e-5_10_epochs_500_steps_gbs_8_wd_0.1_max_grad_norm_1.0_baseline

Checkpoint Used: idm_tars_1.5_7b_frame_pairs_896x896_lr_1e-5_10_epochs_500_steps_gbs_8_wd_0.1_max_grad_norm_1.0_baseline/checkpoint_epoch_9.pt

Model Configuration

• Model Version: TARS 1.5
• Model Size: 7B parameters
• Data Type: Frame pairs
• Learning Rate: 1e-5
• Epochs: 10
• Training Steps: 500
• Global Batch Size: 8
• Weight Decay: 0.1
• Max Gradient Norm: 1.0
• Resolution: 896x896
• Training Data: Baseline

Description

This repository contains the model state dict extracted from the training checkpoint.

Files

• model_state_dict.pt: PyTorch state dictionary containing the model weights
• README.md: This file

Usage

import torch

# Load the model state dict
state_dict = torch.load("model_state_dict.pt", map_location='cpu')

# Use with your model architecture
# model.load_state_dict(state_dict)

Notes

• This model was automatically uploaded using the push_models_to_hf.py script
• The repository name may be truncated if the original model name exceeded HuggingFace's 96-character limit
• Checkpoint extracted from: checkpoint_epoch_9.pt