AI Model Switching via Data Patching and Light Version Files
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing size and diversity of artificial intelligence (AI) models pose a challenge for electronic devices to efficiently switch between different AI models to meet various application requirements.
Innovation Solution
The electronic device compiles multiple AI models into compiled files and generates light version files that include command sets and data patches. These files are stored and loaded during inference, allowing the device to efficiently switch between AI models by generating the necessary data sets from the patches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If full data sets are loaded for each AI model during switching, then model execution accuracy is maintained, but memory usage and switching overhead increase significantly
Solution Approach 1:
The patent segments the complete data set into a base data set and multiple data patches, where each patch represents incremental updates or variations. This allows the system to load only the necessary patch data instead of complete duplicate data sets for each AI model variant, significantly reducing memory consumption while maintaining full model execution capability.
Solution Approach 2:
The patent implements a nested data structure where data patches are embedded within the context of a base data set. The light version file contains references to the base data set location and the patch data, creating a compact nested representation that expands only when needed, thereby reducing the active memory footprint during model switching.
2Reliability
If complete compiled files are loaded for model switching, then model execution reliability is ensured, but switching time and processing overhead increase
Solution Approach 1:
The patent extracts only the essential patch data and command set information from the complete compiled files, separating the variable patch components from the stable base components. This extraction allows the system to load minimal data during switching operations while maintaining full model reliability, as the base data set remains intact and only incremental changes are applied.
Solution Approach 2:
The patent performs preliminary compilation of AI models into light version files that pre-process and organize data patches and command sets. This preliminary action prepares the model switching infrastructure in advance, so that during actual model switching, the system only needs to apply pre-organized patches rather than processing complete compiled files, significantly reducing switching time.
3Adaptability or versatility
If multiple complete compiled files are stored for different AI models, then model diversity and adaptability are maintained, but storage space requirements increase
Solution Approach 1:
The patent merges multiple AI model data sets by identifying and consolidating common base data that appears across multiple models. Instead of storing separate complete compiled files for each model, the system stores a single base data set and multiple lightweight patch files that represent model-specific variations, thereby maintaining model diversity while dramatically reducing total storage requirements.
Solution Approach 2:
The patent uses selective copying where only the differing patch portions are copied and stored for each model variant rather than copying entire data sets. The light version files contain minimal copy information (pointers or references to base data plus patch data), enabling the system to maintain multiple model versions with adaptability while using fraction of the storage space of complete duplicate storage.
Data Source
AI summary
An electronic device is configured to execute instructions: compiling a first AI model and second AI model(s) to a first compiled file and second compiled file(s), respectively, wherein the first compiled file comprises a first data set and a first command set, and the second compiled file(s) comprises second data set(s) and second command set(s); generating light version file(s) for the AI model(s), wherein the light version file(s) comprises the second command set(s) and data patch(es); storing the first compiled file and the light version file(s) to a storage device; loading the first compiled file from the storage device to a memory; loading the light version file(s) from the storage device to the memory; generating the second data set(s) according to the first data set and the data patch(es); and executing the second AI model(s) according to the generated second data set(s) and the second command set(s) in the memory.