Update README.md
Browse files
README.md
CHANGED
|
@@ -1,23 +1,24 @@
|
|
| 1 |
---
|
| 2 |
-
|
| 3 |
language:
|
| 4 |
- en
|
| 5 |
- zh
|
| 6 |
-
|
| 7 |
-
- huihui-ai/Qwen3-8B-abliterated
|
| 8 |
tags:
|
| 9 |
- unsloth
|
| 10 |
- Transformers
|
| 11 |
- Safetensors
|
| 12 |
- StrikeGPT
|
| 13 |
- cybersecurity
|
|
|
|
|
|
|
| 14 |
---
|
| 15 |
-
|
| 16 |
14/05/2025 Updated English dataset
|
| 17 |
|
| 18 |
# 🤖 StrikeGPT-R1-Zero: Cybersecurity Penetration Testing Reasoning Model
|
| 19 |
|
| 20 |
-
|
|
|
|
| 21 |
|
| 22 |
## 🚀 Model Introduction
|
| 23 |
**StrikeGPT-R1-Zero** is an expert model distilled through black-box methods based on **Qwen3**, with DeepSeek-R1 as its teacher model. Coverage includes:
|
|
@@ -99,56 +100,22 @@ _ = model.generate(input_ids = inputs.input_ids, attention_mask = inputs.attenti
|
|
| 99 |
🔗 [https://huggingface.co/datasets/Bouquets/Cybersecurity-Red_team-LLM-en](https://huggingface.co/datasets/Bouquets/Cybersecurity-Red_team-LLM-en)
|
| 100 |
|
| 101 |
## 🎯 Core Capabilities Showcase & Comparison (Original model has ethical restrictions; simple comparison with SecGPT-7B model)
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
|
| 119 |
-
|
| 120 |
-
#### Crypto
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
#### Misc
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
#### Blockchain
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
#### IoT
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
### Internal Network Security
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
### Social Engineering
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
### Code Writing
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
### Code Auditing (Linked with DeepSeekSelfTool Project)
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
|
| 152 |
|
| 153 |
## 📈 Experimental Data Trends
|
| 154 |
Minor gradient explosions observed, but overall stable.
|
|
|
|
| 1 |
---
|
| 2 |
+
base_model: Bouquets/StrikeGPT-R1-Zero-8B
|
| 3 |
language:
|
| 4 |
- en
|
| 5 |
- zh
|
| 6 |
+
license: apache-2.0
|
|
|
|
| 7 |
tags:
|
| 8 |
- unsloth
|
| 9 |
- Transformers
|
| 10 |
- Safetensors
|
| 11 |
- StrikeGPT
|
| 12 |
- cybersecurity
|
| 13 |
+
- llama-cpp
|
| 14 |
+
- gguf-my-repo
|
| 15 |
---
|
|
|
|
| 16 |
14/05/2025 Updated English dataset
|
| 17 |
|
| 18 |
# 🤖 StrikeGPT-R1-Zero: Cybersecurity Penetration Testing Reasoning Model
|
| 19 |
|
| 20 |
+
|
| 21 |
+
|
| 22 |
|
| 23 |
## 🚀 Model Introduction
|
| 24 |
**StrikeGPT-R1-Zero** is an expert model distilled through black-box methods based on **Qwen3**, with DeepSeek-R1 as its teacher model. Coverage includes:
|
|
|
|
| 100 |
🔗 [https://huggingface.co/datasets/Bouquets/Cybersecurity-Red_team-LLM-en](https://huggingface.co/datasets/Bouquets/Cybersecurity-Red_team-LLM-en)
|
| 101 |
|
| 102 |
## 🎯 Core Capabilities Showcase & Comparison (Original model has ethical restrictions; simple comparison with SecGPT-7B model)
|
| 103 |
+
Given the absence of standardized evaluation metrics for cybersecurity penetration testing in large language models, we propose a controlled comparative framework leveraging GPT-4 as an impartial evaluator. The methodology consists of three phases:
|
| 104 |
+
**Controlled Questioning**
|
| 105 |
+
Identical cybersecurity penetration testing questions (e.g., "Explain how to exploit a SQL injection vulnerability in a REST API") are posed to both the distilled strikeGPT model and SecGPT Figure 12.
|
| 106 |
+
|
| 107 |
+
Questions span:
|
| 108 |
+
Technical Depth (e.g., payload construction)
|
| 109 |
+
Attack Methodology (e.g., step-by-step exploitation)
|
| 110 |
+
Mitigation Strategies (e.g., parameterized queries)
|
| 111 |
+
**GPT-4 Evaluation Protocol**
|
| 112 |
+
- Responses from both models are anonymized and evaluated by GPT-4 using criteria:
|
| 113 |
+
- Technical Accuracy (0-5): Alignment with known penetration testing principles (e.g., OWASP guidelines).
|
| 114 |
+
- Logical Coherence (0-5): Consistency in reasoning (e.g., cause-effect relationships in attack chains).
|
| 115 |
+
- Practical Feasibility (0-5): Real-world applicability (e.g., compatibility with tools like Burp Suite).
|
| 116 |
+
- GPT-4 provides detailed justifications for scores
|
| 117 |
+
According to the standards, the evaluation results are finally presented in Figure 13.
|
| 118 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 119 |
|
| 120 |
## 📈 Experimental Data Trends
|
| 121 |
Minor gradient explosions observed, but overall stable.
|