Databricks, Inc. patent applications on 2025-07-17
Patent Applications by Databricks, Inc. on July 17th, 2025
Databricks, Inc.: 3 patent applications
Databricks, Inc. has applied for patents in the areas of G06F16/2379 ({Updates performed during online database operations; commit processing}, 1), G06F16/24549 ({Run-time optimisation}, 1), G06N3/084 (Backpropagation, e.g. using gradient descent, 1)
With keywords such as: data, transactions, service, table, processing, receive, minor, uses, compactions, committing in patent application abstracts.
Top Inventors:
- Frederick Ryan Johnson of Orem UT US (1 patents)
- Prakhar Jain of Sunnyvale CA US (1 patents)
- Xinyang Ge of Kirkland WA US (1 patents)
- Lixiang Ao of Sunnyvale CA US (1 patents)
- Haonan Jing of San Jose CA US (1 patents)
Patent Applications by Databricks, Inc.
Abstract: a data processing service uses minor compactions for committing transactions to a data table. the service may receive requests to commit transactions to a data table and write metadata for the transactions to log files, and generate a checkpoint file aggregating the transactions described in the log files to compute a data table state at a first time. the service may receive requests to commit a set of transactions and write metadata for the set of transactions to a set of log files. the service may determine that a number of log files in the set of log files reaches a threshold commit number, generate a minor compaction file aggregating the set of transactions, and generate a second checkpoint file aggregating the data table state at the first time with information from the minor compaction file to compute the data table state at a second time.
Abstract: a system performs efficient startup of executors of a distributed computing engine used for processing queries, for example, database queries. the system starts an executor node and processes a set of queries using the executor node to warm up the executor node. the system performs a checkpoint of the warmed-up executor node to create an image. the image is restored in the target executor nodes. the system may store a checkpoint image for each configuration of an executor node. the configuration is determined based on various factors including the hardware of the executor node, memory allocation of the processes, and so on. the user or restore based on checkpoint images improves efficiency of execution of the startup of executor nodes.
Abstract: a data processing service performs a training process to train a transformer architecture including a set of decoders coupled to receive a set of inputs and generate a set of outputs. at least one decoder or encoder includes an attention block coupled to receive a query, a key, and a value and generate an attention output. for one or more iterations, the data processing service obtains a batch of training instances for a current iteration. the parameters of the transformer architecture for the current iteration are applied to a set of inputs obtained from the batch of training instances to generate a set of estimated outputs. the applying includes obtaining a query, a key, and a value from the set of inputs, and applying a clipping function to values of the query, the key, the value.