bnadimi/PyraNet-Verilog

PyraNet_Verilog数据集是一个多层次的Verilog代码数据集，包含692,238个Verilog代码样本，每个样本都配有详细的代码描述、编译信息、排名和复杂度等级。该数据集旨在通过增加数据集规模和多样化标签类型来克服现有Verilog数据集的局限性。数据集的结构包括两个主要列：code和description，其中description列包含详细的字典格式信息，如描述、排名、复杂度、编译状态和编译结果。

The PyraNet_Verilog dataset is a multi-layered hierarchical dataset for Verilog code, containing 692,238 Verilog code samples, each with detailed code descriptions. The dataset also includes compile information, rankings, and complexity levels. Compile information is processed through Icarus Verilog, categorizing outputs into two groups: no syntax errors and dependency issue errors. Rankings are evaluated using the GPT-4o-mini model for coding style and efficiency, ranging from 0 to 20. Complexity classification follows the approach in the MEV-LLM paper, categorizing Verilog code into basic, intermediate, advanced, and expert levels. The dataset structure includes two main columns: code and description, where the description column stores detailed information in dictionary format, including description, rank, complexity, compile status, and compile results.