# Model Management The **Model Management** tab in Settings handles downloading, storing, and loading SAM model checkpoints. These checkpoints are required by both the Image Encoder and the Segmentation tool. ## Downloading a Model 1. Open **Geo-SAM Settings** and go to the **Model Management** tab. 2. Locate the model you want in the table — models not yet downloaded are marked in red. 3. Click the **Download** button in the Actions column. 4. Wait for the download to finish. Progress is shown below the table. The model checkpoint is saved to the configured **Model Store Directory** (default: `models/` under the plugin root). ## Available Models The plugin supports models from the SAM, SAM 2, SAM 2.1, and SAM 3 families: | Model | Series | Approximate Size | |---|---|---| | SAM Base | SAM | ~375 MB | | SAM Large | SAM | ~1.2 GB | | SAM2 Tiny | SAM2 | ~39 MB | | SAM2 Small | SAM2 | ~46 MB | | SAM2 Base | SAM2 | ~170 MB | | SAM2 Large | SAM2 | ~898 MB | | SAM2.1 Tiny | SAM2 | ~39 MB | | SAM2.1 Small | SAM2 | ~46 MB | | SAM2.1 Base | SAM2 | ~170 MB | | SAM2.1 Large | SAM2 | ~898 MB | | SAM3 | SAM3 | — | ```{tip} For most use cases, **SAM2.1 Base** offers a good balance between accuracy and speed. Choose **SAM2.1 Tiny** if you need faster encoding with limited GPU memory. ``` ## Deleting a Model Click the **Delete** button in the Actions column to remove a downloaded checkpoint from disk. ## Unloading Models from Memory Click **Unload Current** to release all loaded model sessions from memory. This is useful when you want to free GPU or system memory without deleting the checkpoint files. ## Filtering the Model List Use the filter bar above the table to show: - **All** -- every known model. - **Downloaded** -- models with a local checkpoint file. - **Not Downloaded** -- models that still need to be downloaded. - **In Memory** -- models currently loaded into memory for inference. ## Model Store Directory The default model directory is `models/` inside the plugin folder. You can change it by clicking **Browse** in the Storage section and selecting a different folder. This is useful when you want to share model checkpoints across multiple machines or store them on a larger drive.