Posts

This article explores methods for enabling LLMs to output fixed formats (such as JSON), including the universal prompt approach, post-inference processing solutions (regular expressions), and in-inference processing solutions (dynamic constrained decoding). It analyzes the advantages, disadvantages, and applicable scenarios of each method, and points out the limitations of current research.

This article introduces how to design and implement a short link service, covering key technologies such as high concurrency and security.

This article explores the principles of Java's reflection mechanism and emphasizes its powerful ability to obtain class information at runtime. The article first introduces basic reflection code examples, and then provides an in-depth analysis of the reflection implementation process, including the compilation, loading and running phases. Finally, it is pointed out that since the object creation mechanism in Java involves Class class objects storing specific information of the class, reflection can achieve dynamic acquisition of class information.

This article introduces the physical storage methods of clustered indexes and non-clustered indexes in MySQL. Clustered indexes store the table's index and data together, while non-clustered indexes store them separately. The article further analyzes the performance differences between the two index structures in query and modification operations: the clustered index is faster during query, but may cause B+ tree splitting and merging during modification, while the non-clustered index is relatively less time-consuming.

When there are new modifications to the remote repository and the local repository is not synchronized, continuing to make modifications locally will cause conflicts. Git provides two processing methods: Merge and Rebase. Merge will generate a new submission to summarize the submissions of C and D, but conflicts may occur and need to be resolved manually. Rebase follows linear submission processing, and the history record is cleaner. In terms of usage recommendations, it is safer to use Rebase on personal branches, but be cautious on public branches.