ML Asset Versioning with Serialization for Foundation Models

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Traditional version control systems are inadequate for managing the complex lifecycle of foundation model assets in FMware, including large-scale neural networks, datasets, and dynamic agents, due to their size and frequent changes, lacking tailored management methods and domain-specific operations.

Innovation Solution

A method and system for managing machine learning assets using a version control system that involves retrieving, deserializing, modifying, and serializing assets with customizable serialization and deserialization logic, storing them efficiently in local or external storage, and tracking versions with cryptographic hash verification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional version control systems are used to manage foundation model assets, then basic version tracking is provided, but they cannot handle large-scale models, datasets, and dynamic assets effectively due to size and complexity constraints

Engineering Contradiction:

Improveasset management capabilityVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system segments asset management into distinct components: asset definition (schema), serialization/deserialization logic, version control integration, and storage management. This allows each component to be optimized independently while working together to handle complex foundation model assets effectively

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces serialization and deserialization logic as intermediary components that bridge traditional version control systems and foundation model assets. These intermediaries convert complex asset data into manageable formats, enabling VCS to handle large-scale models and datasets without direct exposure to their complexity

Inventive Principle:

Principle #24Intermediary (Mediator)

2Loss of information

If copies of each modified version are stored to track asset evolution, then complete version history is maintained, but storage space requirements increase significantly

Engineering Contradiction:

Improveversion history completenessVSAvoidstorage space

Core Design Contradiction:

Loss of informationVSQuantity of substance

Solution Approach 1:

The system uses selective copying where only necessary asset versions are stored in full, while other versions are referenced through pointers or metadata. This maintains version history completeness while reducing redundant storage of identical or similar asset copies

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The version control structure nests multiple levels of information: asset data, version metadata, and storage references are nested within each other. This allows the system to store comprehensive version information while keeping the actual asset storage efficient by nesting references within the version control structure

Inventive Principle:

Principle #7Nested doll (Nesting)

3Ease of operation

If traditional asset management approaches are used for foundation model assets, then simple file tracking is provided, but domain-specific operations and tailored management methods are lacking

Engineering Contradiction:

Improveoperation simplicityVSAvoiddomain-specific functionality

Core Design Contradiction:

Ease of operationVSAdaptability or versatility

Solution Approach 1:

The system creates a universal asset management framework that can handle multiple types of foundation model assets (models, datasets, prompts, agents) through a common interface. The schema-based asset definition and standardized serialization/deserialization processes provide universal functionality while accommodating domain-specific requirements through configurable asset types and operations

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20260017048A1Machine learning asset management

Publication Date: 2026.01.15 HUAWEI TECH CO LTD

AI summary

A computer implementable method, an asset developer computer, and a computer readable medium for managing an asset for machine learning are provided. The method comprises retrieving, from a version control system server over a network, an asset for machine learning, deserializing the asset from a serialized asset data format to a deserialized asset data format using a seriazlier/deserializer to generate a deserialized asset, modifying, via at least one processor of an asset developer workstation, the deserialized asset to generate a new version of the asset, serializing the new version of the asset using the serializer/deserializer to generate a serialized new asset, and sending, from the asset developer workstation, the serialized new asset to the version control system server over the network for storage.