HoudiniVexBench: VEX Code Generation Tasks from Houdini 21.0.596

HoudiniVexBench is a benchmark dataset for VEX (Vector Expression Language) code generation and understanding tasks. It contains 86 tasks extracted from Houdini 21.0.596, split into code completion, documentation-to-code, and code explanation categories. The dataset, created by kelvincai and last updated in February 2026, is hosted on HuggingFace.

Use Cases

• Benchmarking code generation models based on the 86 VEX tasks.
• Training models for code completion based on the 61 code_completion tasks.
• Training models for documentation-to-code translation based on the 12 doc_to_code tasks.
• Training models for code explanation based on the 13 code_explanation tasks.
• Evaluating language model performance on domain-specific 3D graphics programming.

Strengths

• Provides a specific benchmark with 86 tasks for a niche programming language.
• Tasks are categorized into three distinct types: code completion, doc_to_code, and code explanation.
• Dataset version 1.0.0 includes an official VEX compilation fix.
• Last updated on 2026-02-11, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect source bias inherent to being extracted solely from Houdini 21.0.596.

Provenance

Source

Extracted from Houdini 21.0.596.

Collection Method

Likely extracted from the software's internal documentation or examples.

Time Range

Based on Houdini version 21.0.596.

Freshness

Last updated 2026-02-11 01:48:02.

Geography

null