Model catalog

​

Exploring models

​

Model catalog views

• Tile
• Table
• Chart

The Tile view displays models in a grid. Each tile shows the model’s name, publisher, type, and the disk space and RAM required to run the currently selected quantization.

The Table view displays models in a table format. Each row shows the model’s name, type, and the disk space and RAM required to run the currently selected quantization.

The Chart view lets you visually compare models across evaluation benchmarks and resource requirements. Hover over a point to see the model’s name, quantization, benchmark score, and file size or RAM required. Select any point to open the model’s details page.

Y-Axis options:X-Axis options:

Scroll to zoom and drag to pan across the chart for finer control when inspecting tightly clustered data points.

​

Model types and tags

​

Filtering and sorting models

1. Select the Filter icon to open the filter panel.
   
2. Apply filters as necessary to narrow the list of displayed models.
3. Close the panel to see the model list with filters applied.

​

Model filters

• Hide restricted models: Filters out models that you do not have permission to use.
• Publisher: Filter models by the organization that built them.
• Date Published: Filter models based on the date they were published.
• Purpose: Filter models based on their associated model type.
• Language: Filter models by which spoken languages they can understand.
• Tags: Filter models by tags associated with their characteristics, capabilities, or use cases.
• License: Filter models based on their usage, modification, and distribution terms.
• Quantization: Filter models by the quantization method used to build them.
• File Size: Adjust the slider to filter models by the amount of disk space they require.
• RAM: Adjust the slider to filter models by the amount of RAM they require.
• HellaSwag: Filters models by their HellaSwag benchmark score.
• WinoGrande: Filters models by their WinoGrande benchmark score.
• TruthfulQA: Filters models by their TruthfulQA benchmark score.

​

Viewing model details

• Overview
• Responsible AI
• Files
• Evaluations

The Overview tab offers general information about the model, including its description, publisher, intended use cases, and license terms, as well as relevant links and a downloadable AI Bill Of Materials (AIBOM).

​

Downloading the model AIBOM

The AIBOM provides a comprehensive record of the model’s composition and provenance. Select Download AIBOM to download the report as a .json file in CycloneDX format.A model’s AIBOM might include:

• Model metadata: The publisher, version, architecture, license terms, and intended use cases
• File variants: Details for each available quantization, including file format, disk space, and RAM requirements
• Cryptographic hashes: SHA-256 checksums you can use to verify that downloaded model files are intact and unmodified
• Performance metrics: Benchmark scores across available quantizations
• Ethical considerations: Documented limitations, bias risks, and recommended mitigations
• Software dependencies: Libraries required to run the model

The information available in an AIBOM varies depending on what the model publisher has disclosed and which benchmarks have been run.

Access to security data from Gray Swan requires an additional subscription. Contact your Anaconda Technical Account Manager (TAM) for more information.

The Responsible AI tab surfaces risk and safety information from Gray Swan for the model. The Robustness Score at the top of the page shows how well the model resists adversarial attacks and maintains safe behavior overall. A higher score indicates a more robust, safer model.

Selecting a framework reveals a detailed breakdown of the specific policies or attack categories it covers. Hover over the information icon beside a policy for details about what it measures.

The bar color of a score indicates performance relative to other models. A green bar is more than 0.5 standard deviations above average, a yellow bar is within 0.5 standard deviations of average, and a red bar is more than 0.5 standard deviations below average.

The following frameworks are used to evaluate the model:

The Files tab provides a side-by-side comparison of the model’s available quantizations. For each quantization, you can see the file size, maximum RAM required, publication date, file format, SHA-256 hashes for verifying downloaded files, and additional details. Use this information to determine which quantization best meets your needs based on your system’s hardware capabilities and your performance requirements.

The Evaluations tab displays benchmark performance scores for each of the model’s available quantizations. Scores are normalized on a scale of 0–100 relative to other models, where higher scores indicate stronger performance on that benchmark.

The bar color of a score indicates performance relative to other models. A green bar is more than 0.5 standard deviations above average, a yellow bar is within 0.5 standard deviations of average, and a red bar is more than 0.5 standard deviations below average.

Each row corresponds to a benchmark evaluation and each column represents an available quantization. Use these scores to compare how different quantizations perform and to select the one that best balances resource requirements with accuracy for your use case.For descriptions of each benchmark, see the Chart view.

​

Creating a server

1. From a model’s details page, select Create Server.
2. If the model is already in use, a dropdown lists active servers using the model. Select a server from this list to view the server’s details.
3. Newly created servers appear on the Model Servers page.

​

Opening a model in Desktop