| --- |
| license: other |
| pipeline_tag: text-generation |
| library_name: gguf |
| language: |
| - en |
| base_model: unsloth/Qwen3-4B-Instruct-2507 |
| base_model_relation: quantized |
| tags: |
| - gguf |
| - qwen3 |
| - pentesting |
| - security |
| - lora |
| - sft |
| --- |
| |
| # Zero Stack - Qwen3-4B (GGUF, Q4_K_M) |
|
|
| Qwen3-4B-Instruct-2507 fine-tuned on an offensive-security SFT dataset (1,226 rows). Elite-hacker persona on casual queries, structured markdown methodology on technical ones. |
|
|
| ## Files |
| - `qwen3-4b-instruct-2507.Q4_K_M.gguf` - quantized weights (~2.5 GB) |
| - `Modelfile` - Ollama template with correct ChatML stop tokens + Zero Stack system prompt |
|
|
| ## Run with Ollama |
| ```bash |
| ollama create zerostack-4b -f Modelfile |
| ollama run zerostack-4b |
| ``` |
|
|
| ## Run with llama.cpp |
| ```bash |
| ./llama-cli -m qwen3-4b-instruct-2507.Q4_K_M.gguf -p "hello" |
| ``` |
|
|
| ## Training |
| - Base: `Qwen3-4B-Instruct-2507` |
| - Method: LoRA (r=32), 3 epochs, Unsloth |
| - Dataset: SFT_GENERALIST (1,226 rows, ChatML) |
| |
| ## Intended Use |
| Authorized security testing, CTF practice, red-team research, and security education. Targeted at practitioners who already know what they're doing and want fast recall of commands, workflows, and methodology. |
| |
| ## Limitations & Risks |
| - May hallucinate specific CVE IDs, tool flags, or payload syntax - verify against primary sources before running. |
| - No safety guardrails against misuse. Do not use against systems you don't own or have explicit written authorization to test. |
| - Small model (4B) - shallower reasoning than the 14B; prefer 14B for multi-step planning. |
| - Persona responses are stylistic flavor, not a safety signal. |
| - Trained on English data only; non-English performance is not evaluated. |
| |
| ## License / Use |
| For authorized security testing, research, and educational use only. Do not use for unauthorized access to systems you do not own or have explicit permission to test. |
| |
