How vext-labs-inc/vext-pentest-7b works
Vext-labs-7B-v1.1 is a 7B language-model checkpoint for autonomous penetration-testing analysis, distributed with a CLI inference script. The documented workflow is to provide scan or tool output, generate model text, and use it for tool-output interpretation, attack planning, vulnerability classification, or remediation guidance. The sampled script loads a tokenizer and causal language model through Transformers; the README also documents vLLM serving. The repository does not show a pentesting agent or tool-execution runtime. Bundled weights and tokenizer files are identified from the tree; their runtime wiring is not shown in the sampled source.
Main entry points: run.py Inference CLI · tokenizer.json Tokenizer · adapter_model.safetensors Model weights
Components 8
Inference
- Inference CLI
run.py - Pentest analysis
run.py
Model assets
- Tokenizer
tokenizer.json - Chat template
chat_template.jinja - Model weights
adapter_model.safetensors - Adapter config
adapter_config.json
Other components
- Analyst
- vLLM serving
Connections 5
- Analyst → Inference CLI: provides prompt
README.md - Inference CLI → Tokenizer: tokenizes input
run.py - Inference CLI → Pentest analysis: generates response
run.py - Pentest analysis → Analyst: returns analysis
README.md - vLLM serving → Model weights: serves checkpoint
README.md