Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,42 @@
+---
+language:
+- code
+tags:
+- python
+- java
+- cpp
+- ai-detection
+- code-analysis
+- temporal-cnn
+- codet5
+metrics:
+- f1: 0.9921
+---
+
+# ai_code_detect
+
+### Architecture
+- **Semantic Engine:** `Salesforce/codet5-base`
+- **Statistical Extraction:** `microsoft/codebert-base-mlm` (Calculates Entropy and Log-Rank across 256 tokens)
+- **Fusion Network:** 1D CNN for temporal feature extraction + Dense Feed-Forward Classifier
+
+### Performance Metrics
+Trained on a polyglot dataset (Python, Java, C++) to prevent single-language overfitting.
+- **Training Validation F1:** 0.9861
+- **Unseen SemEval-2026 Audit (F1):** 0.9921
+- **Overall Accuracy:** 99.20%
+
+### How to use
+To use this model in your own application, download the weights directly from this hub and load them into the custom `TemporalFusionClassifier` architecture.
+
+```python
+from huggingface_hub import hf_hub_download
+import torch
+
+# Download weights
+weights_path = hf_hub_download(repo_id="santh-cpu/ai_code_detect", filename="pytorch_model.bin")
+
+# Load into your architecture
+model = TemporalFusionClassifier(base_model)
+model.load_state_dict(torch.load(weights_path))
+model.eval()