aegis-detect 0.1.0__tar.gz

aegis_detect-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 Anthony Qin
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

aegis_detect-0.1.0/PKG-INFO

@@ -0,0 +1,79 @@
+Metadata-Version: 2.4
+Name: aegis-detect
+Version: 0.1.0
+Summary: AI Code Detection Tool
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: accelerate>=0.20.0
+Requires-Dist: datasets>=2.0.0
+Requires-Dist: huggingface-hub>=0.15.0
+Requires-Dist: matplotlib>=3.5.0
+Requires-Dist: numpy>=1.21.0
+Requires-Dist: openai>=1.0.0
+Requires-Dist: pandas>=1.5.0
+Requires-Dist: peft>=0.4.0
+Requires-Dist: python-dotenv>=0.19.0
+Requires-Dist: safetensors>=0.3.0
+Requires-Dist: scikit-learn>=1.0.0
+Requires-Dist: seaborn>=0.13.2
+Requires-Dist: torch>=2.0.0
+Requires-Dist: tqdm>=4.60.0
+Requires-Dist: transformers>=4.30.0
+Dynamic: license-file
+
+# Aegis: AI Python Code Detection Model
+
+## Overview
+Aegis is a fine-tuned CodeBERT model that classifies AI-generated and human code. CodeBERT has 125 million parameters, but using LoRA (Low-Rank Adaptation), Aegis was efficiently trained locally with only a subset of the original parameters being updated. This project sprouted out of my curiosity of classifying AI and human code off semantic differences. Hence, the dataset (20K samples: 10K AI + 10K Human) was aggressively cleaned to ensure standard formatting and the removal of comments and docstrings. The threshold was set to 0.7 to suggest that were is enough evidence to "pass" the code sample as AI-generated. Predictions by Aegis are not perfect: it's not an end all be all judge. Additionally, tasks where semantic convergence between humans and AI is observed (think LeetCode) are inherently hard to classify. 
+
+## Installation
+
+```bash
+pip install aegis-detect
+```
+
+### CLI Usage
+
+**Supported commands**:
+```bash
+# Predicting using a file
+aegis --file path/to/code.py
+
+# Predicting using text
+aegis --text "def add(a, b):\n    return a + b"
+
+# JSON output
+aegis --file path/to/code.py --json > result.json
+
+# Setting a threshold for AI classification 
+aegis --file path/to/code.py --threshold 0.7
+
+# Help
+aegis --help
+```
+
+**Notes**:
+- On first run, the model adapter is downloaded from the Hugging Face repo [anthonyq7/aegis](https://huggingface.co/anthonyq7/aegis) and cached under `~/.aegis/models`.
+- Internet access is required on the first run; subsequent runs use local cache. 
+- The CLI prints the predicted label and probabilities for human and AI.
+
+## Key Results
+
+### Model Performance
+- **Accuracy**: 85.10%
+- **Precision**: 83.37%
+- **Recall**: 87.70%
+- **F1-Score**: 85.48%
+
+### Confusion Matrix
+![Alt text](model/results/confusion_matrix.png)
+
+### Attention Heatmap
+![Alt text](model/results/attention_weights.png)
+
+## Contact
+**Email**: a.j.qin@wustl.edu
+
+## License
+This project is licensed under the [MIT License](LICENSE).

aegis_detect-0.1.0/README.md

@@ -0,0 +1,55 @@
+# Aegis: AI Python Code Detection Model
+
+## Overview
+Aegis is a fine-tuned CodeBERT model that classifies AI-generated and human code. CodeBERT has 125 million parameters, but using LoRA (Low-Rank Adaptation), Aegis was efficiently trained locally with only a subset of the original parameters being updated. This project sprouted out of my curiosity of classifying AI and human code off semantic differences. Hence, the dataset (20K samples: 10K AI + 10K Human) was aggressively cleaned to ensure standard formatting and the removal of comments and docstrings. The threshold was set to 0.7 to suggest that were is enough evidence to "pass" the code sample as AI-generated. Predictions by Aegis are not perfect: it's not an end all be all judge. Additionally, tasks where semantic convergence between humans and AI is observed (think LeetCode) are inherently hard to classify. 
+
+## Installation
+
+```bash
+pip install aegis-detect
+```
+
+### CLI Usage
+
+**Supported commands**:
+```bash
+# Predicting using a file
+aegis --file path/to/code.py
+
+# Predicting using text
+aegis --text "def add(a, b):\n    return a + b"
+
+# JSON output
+aegis --file path/to/code.py --json > result.json
+
+# Setting a threshold for AI classification 
+aegis --file path/to/code.py --threshold 0.7
+
+# Help
+aegis --help
+```
+
+**Notes**:
+- On first run, the model adapter is downloaded from the Hugging Face repo [anthonyq7/aegis](https://huggingface.co/anthonyq7/aegis) and cached under `~/.aegis/models`.
+- Internet access is required on the first run; subsequent runs use local cache. 
+- The CLI prints the predicted label and probabilities for human and AI.
+
+## Key Results
+
+### Model Performance
+- **Accuracy**: 85.10%
+- **Precision**: 83.37%
+- **Recall**: 87.70%
+- **F1-Score**: 85.48%
+
+### Confusion Matrix
+![Alt text](model/results/confusion_matrix.png)
+
+### Attention Heatmap
+![Alt text](model/results/attention_weights.png)
+
+## Contact
+**Email**: a.j.qin@wustl.edu
+
+## License
+This project is licensed under the [MIT License](LICENSE).

aegis_detect-0.1.0/aegis/__init__.py

@@ -0,0 +1,3 @@
+__version__ = "0.1.0"
+
+from aegis.predictor import Predictor

aegis_detect-0.1.0/aegis/cli.py

@@ -0,0 +1,76 @@
+"""CLI for Aegis"""
+
+import argparse
+import json
+from argparse import RawDescriptionHelpFormatter
+from pathlib import Path
+from typing import Optional
+
+from transformers import logging as hf_logging
+
+from aegis.predictor import Predictor
+
+hf_logging.set_verbosity_error()
+
+def get_code_input(text: Optional[str], file_path: Optional[str]) -> str:
+    if text:
+        return text
+
+    file = Path(file_path).expanduser()
+    if not file.exists():
+        raise SystemExit(f"Error: File not found: {file_path}")
+    return file.read_text(encoding="utf-8")
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(
+        prog="aegis",
+        description="Detect whether code is human-written or AI-generated",
+        formatter_class=RawDescriptionHelpFormatter,
+        epilog=(
+            "Examples:\n"
+        "  aegis --file path/to/code.py\n"
+        "  aegis --text 'def add(a, b): return a + b'\n"
+        "  aegis --file script.py --json > result.json\n"
+        "  aegis --threshold 0.7\n"
+        )
+    )
+
+    #creates a mutually exclusive input group where only either --text or --file is allowed
+    input_group = parser.add_mutually_exclusive_group(required=True)
+    input_group.add_argument("--text", type=str, help="Code snippet as text (quote if it contains spaces or special characters).")
+    input_group.add_argument("--file", type=str, help="Path to code file")
+
+    #output format
+    parser.add_argument("--json", action="store_true", help="Output results as JSON")
+
+    #allows users to adjust a threshold
+    parser.add_argument("--threshold", type=float, help="Custom threshold to classify as AI")
+
+    #parses args
+    args = parser.parse_args()
+
+    #gets the code to analyze
+    code = get_code_input(args.text, args.file)
+
+    #loads model and makes a prediction
+    if args.threshold is None:
+        predictor = Predictor()
+    elif args.threshold:
+        predictor = Predictor(threshold=args.threshold)
+    elif args.threshold == 0:
+        predictor = Predictor(threshold=0)
+    else:
+        raise ValueError("Threshold must be a float between 0 and 1")
+
+
+    result = predictor.predict(code)
+
+    if args.json:
+        print(json.dumps(result, ensure_ascii=False))
+    else:
+        print(f"Prediction: {result['prediction']} | human={result['human']} | ai={result['ai']}")
+
+
+if __name__ == "__main__":
+    main()

aegis_detect-0.1.0/aegis/predictor.py

@@ -0,0 +1,91 @@
+"""Core logic for the CLI interface"""
+
+from pathlib import Path
+from typing import Optional
+
+import torch
+from huggingface_hub import snapshot_download
+from peft import PeftModel
+from transformers import AutoModelForSequenceClassification, AutoTokenizer
+
+
+class Predictor:
+
+    def __init__(self, threshold: Optional[float] = None, model_name="anthonyq7/aegis"):
+
+        if threshold:
+            if not (0.0 <= threshold <= 1.0):
+                raise ValueError("Threshold must be a float between 0 and 1")
+
+        if threshold is None:
+            self.threshold = 0.7
+        else:
+            self.threshold = float(threshold)
+
+        self.model_path = self._get_or_download_model(model_name)
+        self._load_model()
+
+    def _get_or_download_model(self, model_name):
+        cached_dir = Path.home() / ".aegis" / "models"
+        cached_dir.mkdir(parents=True, exist_ok=True)
+
+        if not any(cached_dir.iterdir()):
+            try:
+                snapshot_download(
+                    repo_id=model_name,
+                    local_dir=cached_dir
+                )
+                print("\u2713 Model downloaded successfully!")
+            except Exception as e:
+                print(f"\u2717 Error downloading mode: {e}")
+                print(f"Make sure the model exists at huggingface.co/{model_name}")
+                raise
+
+        return cached_dir
+
+    def _load_model(self):
+
+        print("Loading model...")
+
+        self.tokenizer = AutoTokenizer.from_pretrained(str(self.model_path))
+        base_model = AutoModelForSequenceClassification.from_pretrained("microsoft/codebert-base", num_labels=2)
+        self.model = PeftModel.from_pretrained(base_model, str(self.model_path))
+        self.model.eval()
+
+        print("\u2713 Model loaded!")
+
+    def predict(self, code):
+
+        if not code or not code.strip():
+            raise ValueError("Code cannot be empty")
+
+        inputs = self.tokenizer(
+            code,
+            truncation=True,
+            padding="max_length",
+            max_length=512,
+            return_tensors="pt"
+        )
+
+        with torch.no_grad():
+            outputs = self.model(**inputs)
+            probs = torch.nn.functional.softmax(outputs.logits, dim=-1).cpu().numpy().tolist()[0]
+
+        human_prob = f"{float(probs[0]):.4f}"
+        ai_prob = f"{float(probs[1]):.4f}"
+
+        return {
+            "human": human_prob,
+            "ai": ai_prob,
+            "prediction": "ai-generated" if float(ai_prob) >= self.threshold else "human"
+        }
+
+
+
+
+
+
+
+
+
+

aegis_detect-0.1.0/aegis_detect.egg-info/PKG-INFO

@@ -0,0 +1,79 @@
+Metadata-Version: 2.4
+Name: aegis-detect
+Version: 0.1.0
+Summary: AI Code Detection Tool
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: accelerate>=0.20.0
+Requires-Dist: datasets>=2.0.0
+Requires-Dist: huggingface-hub>=0.15.0
+Requires-Dist: matplotlib>=3.5.0
+Requires-Dist: numpy>=1.21.0
+Requires-Dist: openai>=1.0.0
+Requires-Dist: pandas>=1.5.0
+Requires-Dist: peft>=0.4.0
+Requires-Dist: python-dotenv>=0.19.0
+Requires-Dist: safetensors>=0.3.0
+Requires-Dist: scikit-learn>=1.0.0
+Requires-Dist: seaborn>=0.13.2
+Requires-Dist: torch>=2.0.0
+Requires-Dist: tqdm>=4.60.0
+Requires-Dist: transformers>=4.30.0
+Dynamic: license-file
+
+# Aegis: AI Python Code Detection Model
+
+## Overview
+Aegis is a fine-tuned CodeBERT model that classifies AI-generated and human code. CodeBERT has 125 million parameters, but using LoRA (Low-Rank Adaptation), Aegis was efficiently trained locally with only a subset of the original parameters being updated. This project sprouted out of my curiosity of classifying AI and human code off semantic differences. Hence, the dataset (20K samples: 10K AI + 10K Human) was aggressively cleaned to ensure standard formatting and the removal of comments and docstrings. The threshold was set to 0.7 to suggest that were is enough evidence to "pass" the code sample as AI-generated. Predictions by Aegis are not perfect: it's not an end all be all judge. Additionally, tasks where semantic convergence between humans and AI is observed (think LeetCode) are inherently hard to classify. 
+
+## Installation
+
+```bash
+pip install aegis-detect
+```
+
+### CLI Usage
+
+**Supported commands**:
+```bash
+# Predicting using a file
+aegis --file path/to/code.py
+
+# Predicting using text
+aegis --text "def add(a, b):\n    return a + b"
+
+# JSON output
+aegis --file path/to/code.py --json > result.json
+
+# Setting a threshold for AI classification 
+aegis --file path/to/code.py --threshold 0.7
+
+# Help
+aegis --help
+```
+
+**Notes**:
+- On first run, the model adapter is downloaded from the Hugging Face repo [anthonyq7/aegis](https://huggingface.co/anthonyq7/aegis) and cached under `~/.aegis/models`.
+- Internet access is required on the first run; subsequent runs use local cache. 
+- The CLI prints the predicted label and probabilities for human and AI.
+
+## Key Results
+
+### Model Performance
+- **Accuracy**: 85.10%
+- **Precision**: 83.37%
+- **Recall**: 87.70%
+- **F1-Score**: 85.48%
+
+### Confusion Matrix
+![Alt text](model/results/confusion_matrix.png)
+
+### Attention Heatmap
+![Alt text](model/results/attention_weights.png)
+
+## Contact
+**Email**: a.j.qin@wustl.edu
+
+## License
+This project is licensed under the [MIT License](LICENSE).

aegis_detect-0.1.0/aegis_detect.egg-info/SOURCES.txt

@@ -0,0 +1,12 @@
+LICENSE
+README.md
+pyproject.toml
+aegis/__init__.py
+aegis/cli.py
+aegis/predictor.py
+aegis_detect.egg-info/PKG-INFO
+aegis_detect.egg-info/SOURCES.txt
+aegis_detect.egg-info/dependency_links.txt
+aegis_detect.egg-info/entry_points.txt
+aegis_detect.egg-info/requires.txt
+aegis_detect.egg-info/top_level.txt

aegis_detect-0.1.0/aegis_detect.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

aegis_detect-0.1.0/aegis_detect.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+aegis = aegis.cli:main

aegis_detect-0.1.0/aegis_detect.egg-info/requires.txt

@@ -0,0 +1,15 @@
+accelerate>=0.20.0
+datasets>=2.0.0
+huggingface-hub>=0.15.0
+matplotlib>=3.5.0
+numpy>=1.21.0
+openai>=1.0.0
+pandas>=1.5.0
+peft>=0.4.0
+python-dotenv>=0.19.0
+safetensors>=0.3.0
+scikit-learn>=1.0.0
+seaborn>=0.13.2
+torch>=2.0.0
+tqdm>=4.60.0
+transformers>=4.30.0

aegis_detect-0.1.0/aegis_detect.egg-info/top_level.txt

@@ -0,0 +1 @@
+aegis

aegis_detect-0.1.0/pyproject.toml

@@ -0,0 +1,59 @@
+[project]
+name = "aegis-detect"
+version = "0.1.0"
+description = "AI Code Detection Tool"
+readme = "README.md"
+requires-python = ">=3.10" 
+dependencies = [
+    "accelerate>=0.20.0",
+    "datasets>=2.0.0",
+    "huggingface-hub>=0.15.0",
+    "matplotlib>=3.5.0",     
+    "numpy>=1.21.0",
+    "openai>=1.0.0",
+    "pandas>=1.5.0",
+    "peft>=0.4.0",
+    "python-dotenv>=0.19.0",
+    "safetensors>=0.3.0",
+    "scikit-learn>=1.0.0",
+    "seaborn>=0.13.2",
+    "torch>=2.0.0",
+    "tqdm>=4.60.0",
+    "transformers>=4.30.0",
+]
+
+[project.scripts]
+aegis = "aegis.cli:main"
+
+[build-system]
+requires = ["setuptools>=68", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools]
+packages = ["aegis"]
+
+[tool.uv]
+dev-dependencies = [
+    "ruff>=0.6.0"
+]
+
+[tool.ruff]
+exclude = [
+    ".git",
+    "__pycache__",
+    ".venv",
+    "venv",
+    "node_modules",
+    "build",
+    "dist",
+    "*.egg-info",
+    "saved_models",
+    "data/raw",
+]
+
+[tool.ruff.lint]
+select = ["E", "W", "F", "I", "N"]
+ignore = ["E501"]
+
+[tool.ruff.lint.per-file-ignores]
+"__init__.py" = ["F401"]

aegis_detect-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+