Assemblage Deephistory: 73,610 Binaries Across 248 Projects

73,610 binaries spanning 248 open-source projects, compiled with multiple compilers and optimization levels for Linux and Windows. The dataset includes multi-year version histories and is linked to 329 CVEs. It was uploaded by author changliu8541 to Hugging Face, with a fix noted in May 2026.

Use Cases

• Training vulnerability detection models based on binary features across compiler versions.
• Analyzing the evolution of software security properties over multi-year version histories.
• Studying the impact of different compilers and optimization levels on binary characteristics.
• Benchmarking binary similarity and function matching algorithms on a large-scale dataset.

Strengths

• 73,610 binaries provide a substantial corpus for analysis.
• Covers 248 distinct open-source projects for diversity.
• Includes 329 CVEs for ground-truth vulnerability linkage.
• Spans multiple compilers, optimization levels, and operating systems (Linux and Windows).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.
• A bug in function name matching was reported, requiring users to verify the fixed version.

Provenance

Source

Hugging Face (author: changliu8541)

Collection Method

Assembled from 248 open-source projects.

Time Range

Multi-year version histories.

Freshness

Last updated 2026-05-06 07:15:00; freshness should be verified.