💰 mAGIcan Dev Log
😖 Everyone was waiting for this! mAGIcan updates and news on its development since the release of the DEMO video!
😕 We have finalized and improved the consensus of nodes when restarting.
😕 We moved the models to temporary nodes, which allowed us to significantly save disposal and increase productivity without increasing costs.
😕 Changed the formats of returned messages in process status. Now, instead of strings, only valid JSON is used, which greatly simplifies data processing and analysis.
This change made the system more transparent and easier to use
😕 We tested the functionality of dividing models into basic “cubes”. Although this method was found to be 100 ms slower, its potential benefits for system modularity and scalability are clear.
We continue research in this direction to make the most of the capabilities of this technology.
😕 We improved the deployment scripts for different stands, which made it possible to speed up and simplify the process of deploying updates.
This is an important improvement that will have a positive impact on the efficiency and quality of the services we provide.
😕Finally, we have improved the work with NERs in BERT and LDA models.
These improvements have improved the accuracy and efficiency of natural language processing, which is especially important for our analytics tools. 🚀
😖 Everyone was waiting for this! mAGIcan updates and news on its development since the release of the DEMO video!
TL;DR:
We've optimized our servers to improve performance and simplify data processing by improving message format and AI model accuracy. We also accelerated the process of deploying updates on different platforms.
😕 We have finalized and improved the consensus of nodes when restarting.
😕 We moved the models to temporary nodes, which allowed us to significantly save disposal and increase productivity without increasing costs.
😕 Changed the formats of returned messages in process status. Now, instead of strings, only valid JSON is used, which greatly simplifies data processing and analysis.
This change made the system more transparent and easier to use
😕 We tested the functionality of dividing models into basic “cubes”. Although this method was found to be 100 ms slower, its potential benefits for system modularity and scalability are clear.
We continue research in this direction to make the most of the capabilities of this technology.
😕 We improved the deployment scripts for different stands, which made it possible to speed up and simplify the process of deploying updates.
This is an important improvement that will have a positive impact on the efficiency and quality of the services we provide.
😕Finally, we have improved the work with NERs in BERT and LDA models.
These improvements have improved the accuracy and efficiency of natural language processing, which is especially important for our analytics tools. 🚀