datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
ComPile
Dataset Card for ComPile: A Large IR Dataset from Production Sources
Changelog
Release
Programming Languages
Description
v1.0
C/C++, Rust, Swift, Julia
Fine Tuning-scale dataset of 602GB of deduplicated LLVM (bitcode) IR
Dataset Summary
ComPile contains over 2.7TB of permissively-licensed source code compiled to (textual) LLVM
intermediate representation (IR) covering C/C++, Rust, Swift, and Julia.
The dataset was created by hooking into LLVM… See the full description on the dataset page: https://huggingface.co/datasets/llvm-ml/ComPile.maplm_v2MAPLM: A Real-World Large-Scale Vision-Language Benchmark for Map and Traffic Scene Understandingllvm-apr-benchmark
LLVM APR Benchmark: A Large-Scale Automated Program Repair Benchmark of Real-World LLVM Middle-End Bugs
GitHub (We only accept pull requests from GitHub)
Hugging Face Mirror
Hugging Face Leaderboard
Evaluation Result Submission
Motivation
The compiler is a critical infrastructure in the software development. The LLVM compiler infrastructure is widely used in both academia and industry. However, due to its inherent complexity, the LLVM compiler still contains many bugs… See the full description on the dataset page: https://huggingface.co/datasets/dtcxzyw/llvm-apr-benchmark.x86-to-llvm-o2-updatedx86-to-llvm-o2llvm-apr-benchmark-submissionsCompRealVul_LLVM
CompRealVul_LLVM Dataset
Dataset Summary
CompRealVul_LLVM is the LLVM IR (Intermediate Representation) version of the CompRealVul_C dataset. This version is designed specifically for training and evaluating machine learning models on the task of binary vulnerability detection in a setting that closely mimics how models are used in practice — operating on the compiled representation of code rather than raw source code.
Each function in this dataset was compiled from C… See the full description on the dataset page: https://huggingface.co/datasets/CCompote/CompRealVul_LLVM.autonomous-llvm-mlir-compiler-suite
⚡ Autonomous Compiler Internals, LLVM & MLIR Architecture Suite (2026)
A Production-Grade, Verifiable Synthetic Corpus for Training Frontier Coding Models (Qwen 3.8, DeepSeek-V3, Llama 3.3)
⚡ Overview & Industry Problem
Modern deep learning accelerators, custom ASICs, and high-performance computing clusters demand specialized, autonomous compilation infrastructure: LLVM IR custom passes, SSA dominance frontiers, Chaitin-Briggs graph coloring… See the full description on the dataset page: https://huggingface.co/datasets/beatsprom/autonomous-llvm-mlir-compiler-suite.cisb-bench-llvm-O2
CISB-Bench LLVM-O2: Assembly Transpilation Benchmark
Overview
This dataset is derived from the Compiler-Introduced Security Bugs (CISB) benchmark,
compiled with LLVM/Clang-17 at -O2 optimization level. It supports bidirectional
assembly transpilation (x86-64 ↔ AArch64) and is designed for evaluating transpilation
models on security-critical code.
Dataset Structure
Column
Description
task_name
Full task identifier including flag (e.g.… See the full description on the dataset page: https://huggingface.co/datasets/Akirayasha/cisb-bench-llvm-O2.humaneval_x_llvm_wasm
Dataset Card for "humaneval_x_llvm_wasm"
More Information needed
maplm_v1_5MAPLM: A Real-World Large-Scale Vision-Language Benchmark for Map and Traffic Scene UnderstandingJuliet_LLVM
Juliet_LLVM Dataset
Dataset Summary
Juliet_LLVM is a dataset of compiled C functions from the Juliet Test Suite (as organized in the GitHub repository
), translated into LLVM Intermediate Representation (IR) after pre-process phase. It is designed for training and evaluating machine learning models on the task of binary vulnerability detection. Each function is labeled as either vulnerable or non-vulnerable and is presented in an architecture-agnostic, semantically rich… See the full description on the dataset page: https://huggingface.co/datasets/CCompote/Juliet_LLVM.buzz_sources_152_llvmexebench_loop_optimized_llvm_ir_1sthalftrain_synth_rich_io_filtered_0_llvm_extract_func_ir_assembly_O2_llvm_diff_sample_100mbxp_llvm_wasm
Dataset Card for "hmbxp_llvm_wasm"
More Information needed
x86-to-llvm-o0
x86-to-llvm-o0
This dataset contains pairs of x86 assembly and LLVM IR compiled at optimization level O0, along with file identifiers.It is designed for research in neurosymbolic transpilation, compiler translation, and program analysis.
Dataset Summary
Rows: ~1,022,891
Columns:
x86: x86 assembly code compiled with Clang at -O0
llvm: corresponding LLVM IR at -O0
file: identifier for the original source file
This dataset allows learning mappings between… See the full description on the dataset page: https://huggingface.co/datasets/adpretko/x86-to-llvm-o0.lifted-llvmir-to-src-instruct-datasetcombined_LLVM
CombinedVuln_LLVM Dataset
Dataset Summary
Combined_LLVM is a unified dataset for binary vulnerability detection that merges two sources:
CompRealVul_LLVM, based on real-world C vulnerabilities
Juliet_LLVM, based on the Juliet Test Suite
All functions are provided in LLVM Intermediate Representation (IR) format and labeled as vulnerable or non-vulnerable. This dataset is ideal for training and benchmarking machine learning models that aim to detect vulnerabilities in… See the full description on the dataset page: https://huggingface.co/datasets/CCompote/combined_LLVM.llvm-ir-loop-optimizedllvm-ir-smallexebench_train_synth_rich_io_filtered_llvm_extract_func_ir_assembly_O2_llvm_diffllvm-ir-v2.0lifted-llvmir-to-src-llvmir-instruct-datasetllvm-loop-optimizedexebench_llvm_o3