bytesbrains/hunter-crypto-7b
The bytesbrains/hunter-crypto-7b is a 7.6 billion parameter model developed by Naderu, a BytesBrains Pte. Ltd. venture, fine-tuned from Qwen/Qwen2.5-Coder-7B-Instruct. This specialized model excels at identifying weaknesses in cryptographically weak or misconfigured constructions and generating runnable attack scripts (PyCryptodome/SageMath) to recover plaintext or flags. It is designed for offline use on Apple Silicon, focusing on the cryptography aspect of penetration testing and CTF work.
Loading preview...
Overview
hunter-crypto-7b is a 7.6 billion parameter model developed by Naderu, a BytesBrains Pte. Ltd. venture, specifically designed for cryptography-focused penetration testing and CTF challenges. Fine-tuned from Qwen/Qwen2.5-Coder-7B-Instruct using QLoRA, this model identifies cryptographic weaknesses and generates executable attack scripts (PyCryptodome/SageMath) to recover plaintext or flags. It operates fully offline on Apple Silicon, ensuring engagement data remains local.
Key Capabilities
- Specialized Cryptography Attacks: Focuses on identifying and exploiting weaknesses in cryptographically weak or misconfigured ciphers, RSA keys, and oracles.
- Script Generation: Produces runnable attack scripts that, when executed, recover the target plaintext or flag.
- Offline Operation: Designed for full offline execution on Apple Silicon, enhancing security and data privacy for authorized security work.
- High Solve Rate: Achieved a 1.000 solve rate on in-distribution challenges and 0.972 on independent-method challenges during evaluation.
Intended Use Cases
- Authorized Penetration Testing: Assisting security professionals in assessing the cryptography corner of systems they have explicit permission to test.
- CTF Competitions: Providing a tool for participants to solve cryptography challenges.
- Security Education: Aiding in the learning and understanding of cryptographic vulnerabilities.
Limitations
- This is a small, narrow model focused on weak crypto; it is not a general cryptanalysis engine and may be incorrect on complex or out-of-distribution problems. Generated scripts require sandboxed execution and verification.