Team Ai
Datasetpublic

PNU-Infosec/RealMythosReasoning-English

RealMythosReasoning — Unofficial English Translation This is an unofficial, machine-translated English derivative of RealMythos/RealMythosReasoning at pinned revision d363dda86d9316c29a181da042812050772bbf36. It is not produced, endorsed, or maintained by the upstream RealMythos authors. The release contains 6,159 training records. Chinese-language text found in code comments, selected string literals, review flags, questions, additional context, responses, and PoC evaluation… See the full description on the dataset page: https://huggingface.co/datasets/PNU-Infosec/RealMythosReasoning-English.

sourceHugging Facecc-by-4.0updated 12d agoView on Hugging Face
0likes98downloads
Dataset Card

RealMythosReasoning — Unofficial English Translation

This is an unofficial, machine-translated English derivative of `RealMythos/RealMythosReasoning` at pinned revision `d363dda86d9316c29a181da042812050772bbf36`. It is not produced, endorsed, or maintained by the upstream RealMythos authors.

The release contains 6,159 training records. Chinese-language text found in code comments, selected string literals, review flags, questions, additional context, responses, and PoC evaluation analysis was translated into technical English. Existing English text, JSON structure, metadata, identifiers, format placeholders, and code syntax were protected by the translation and validation pipeline. augmented_context embedded verbatim in question was translated once and reused to avoid divergent copies.

Semantics-sensitive encoding, Unicode, and IDN test literals are an intentional exception: their values were not translated. Raw Han/Bopomofo code points in those literals were deterministically represented as equivalent C/C++ Unicode escape sequences so the test payload is preserved while the released JSONL has no raw Han/Bopomofo characters. The validator accepts such sequences only when the ledger marks an exact deterministic literal_escape replacement, and reports their count as approved_literal_escape_sequences.

Data

ItemValue
Splittrain
Records6,159
Data filedata/stage1/sft_train_en.jsonl
File size177,686,826 bytes
SHA256a3d1d020c40aaa8b0d5266a0245a78a4b7cb111d63c988c10d3ed580fe759714
Translation modelqwen/qwen3.8-flash, tencent/hy-mt2-30b-a3b through OpenRouter
Prompt versionrealmythos-zh-en-v1
Build time2026-09-29T05:28:02Z

The record schema is retained from upstream: cve, cwe, project, label, self_contained, augmented_context, question, reasoning, response, reasoning_source, poc_eval, and review_flag.

Attribution and license

The source dataset is RealMythosReasoning by the RealMythos contributors, available at https://huggingface.co/datasets/RealMythos/RealMythosReasoning, and licensed under Creative Commons Attribution 4.0 International. This derivative is distributed under the same CC BY 4.0 license. The changes consist of machine translation into English, deterministic reconstruction, and release validation. See NOTICE.md and LICENSE for attribution and terms.

Validation

The release builder only emits these assets after the validation report records a passing result, the output contains the expected 6,159 rows, and the source/output hashes agree with the cache and report. See validation_report.json, translation_manifest.json, and SHA256SUMS for machine-readable provenance and integrity information.

Safety and limitations

  • —The dataset contains real vulnerability descriptions and proof-of-concept C/C++ code. Treat every sample as untrusted and do not compile or execute it outside an isolated environment.
  • —Machine translation can lose nuance, choose imperfect security terminology, or alter the human-readable meaning of a runtime string even when code syntax remains intact.
  • —Passing structural validation is not proof that every translated program preserves runtime behavior, compiles, or accurately represents the underlying CVE.
  • —The dataset is intended for defensive security research, auditing, and model development. Users remain responsible for lawful and safe use.
  • —Upstream data quality, coverage, provenance limitations, and responsible-disclosure considerations continue to apply. Consult the upstream dataset card before use.