Jarvis-7B-Unified-Coder
Hybrid language model for coding and cybersecurity
1 min readupdated
The model is a merge of four specialized versions based on the Qwen2.5-Coder-7B-Instruct architecture using the TIES Consensus method. The tool combines precise code editing capabilities, removed annoying refusals (abliterated), and cybersecurity analysis without unnecessary restrictions. The model is useful for developers, engineers, and security specialists for complex multi-level programming and code auditing tasks.
Key features
- Base on Qwen2.5-Coder-7B-Instruct preserving the architecture
- Uncensored and without refusals for cybersecurity and engineering requests
- Microsoft SOTA Code Editing technologies included for deep code refactoring
- Specialization in low-level system programming and vulnerability analysis
What you can do
- Perform complex multi-step code refactoring and maintain syntactic integrity
- Solve information security, auditing, and exploit analysis tasks without refusals
- Use structured tool-calling in JSON format for agents
Example
The model runs through a standard ChatML tokenizer and supports Safetensors formats for integration into familiar LLM inference pipelines.
Read more
DetailsMerge method: TIES Consensus
Base anchor: Qwen/Qwen2.5-Coder-7B-Instruct
Format: Safetensors + ChatML Tokenizer
Density: 0.65, data type: float16










Comments