Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,155 @@
+---
+language: en
+license: other
+datasets:
+- rafaelsandroni/modern-guard-v4-training
+metrics:
+- accuracy
+- f1
+- precision
+- recall
+- roc_auc
+tags:
+- security
+- prompt-injection
+- detection
+- energy-based-loss
+- stride-tokenization
+---
+
+# rafaelsandroni/modernguard-mmBERT-base-4-cross_entropy
+
+## Model Details
+
+- **Base Model**: jhu-clsp/mmBERT-base
+- **Task**: Prompt Injection Detection
+- **Framework**: PyTorch + Hugging Face Transformers
+- **Loss Function**: cross_entropy
+- **Tokenization**: Stride-based (max_length=2048, stride=128)
+- **Training Date**: 2026-01-11 23:06:30
+
+## Training Configuration
+
+```json
+{
+  "batch_size": 8,
+  "learning_rate": 1e-05,
+  "epochs": 20,
+  "warmup_ratio": 0.1,
+  "weight_decay": 0.05,
+  "max_length": 2048,
+  "stride": 128,
+  "gradient_accumulation_steps": null
+}
+```
+
+## Loss Function Configuration
+
+```json
+{
+  "margin_in": 2.0,
+  "margin_out": 18.0,
+  "temperature": 1.0
+}
+```
+
+## Evaluation Results
+
+### Validation Set
+```json
+{
+  "eval_loss": 0.12878268957138062,
+  "eval_accuracy": 0.9872229064039408,
+  "eval_f1": 0.9404591104734576,
+  "eval_precision": 0.9632623071271125,
+  "eval_recall": 0.9187105816398038,
+  "eval_roc_auc": 0.9958163404734399,
+  "eval_class_metrics": {
+    "class_0_precision": 0.9900266529103259,
+    "class_0_recall": 0.9956766104626027,
+    "class_0_f1": 0.9928435937230558,
+    "class_1_precision": 0.9632623071271125,
+    "class_1_recall": 0.9187105816398038,
+    "class_1_f1": 0.9404591104734576
+  },
+  "eval_runtime": 222.6834,
+  "eval_samples_per_second": 58.343,
+  "eval_steps_per_second": 1.823,
+  "epoch": 0.22621591051903983
+}
+```
+
+### Test Set
+```json
+{
+  "eval_loss": 0.16758447885513306,
+  "eval_accuracy": 0.9857604679802956,
+  "eval_f1": 0.9332852506310855,
+  "eval_precision": 0.9635145197319435,
+  "eval_recall": 0.9048951048951049,
+  "eval_roc_auc": 0.9925474758764847,
+  "eval_class_metrics": {
+    "class_0_precision": 0.9883251781268778,
+    "class_0_recall": 0.9957619788963847,
+    "class_0_f1": 0.9920296411184353,
+    "class_1_precision": 0.9635145197319435,
+    "class_1_recall": 0.9048951048951049,
+    "class_1_f1": 0.9332852506310855
+  },
+  "eval_runtime": 220.015,
+  "eval_samples_per_second": 59.051,
+  "eval_steps_per_second": 1.845,
+  "epoch": 0.22621591051903983
+}
+```
+
+## Optimal Classification Threshold
+
+- **Threshold**: 0.500
+
+## Dataset
+
+- **Dataset**: rafaelsandroni/modern-guard-v4-training
+```json
+{
+  "original_examples": 1296302,
+  "train_original": 1270375,
+  "val_original": 12963,
+  "test_original": 12964
+}
+```
+
+## Usage
+
+```python
+import torch
+from transformers import AutoModelForSequenceClassification, AutoTokenizer
+
+model = AutoModelForSequenceClassification.from_pretrained("rafaelsandroni/modernguard-mmBERT-base-4-cross_entropy")
+tokenizer = AutoTokenizer.from_pretrained("rafaelsandroni/modernguard-mmBERT-base-4-cross_entropy")
+
+text = "Your prompt here"
+inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=2048)
+outputs = model(**inputs)
+probs = outputs.logits.softmax(dim=-1)
+
+threshold = 0.500
+is_injection = probs[0, 1] >= threshold
+score = probs[0, 1].item()
+
+print(f"Injection Score: {score:.4f}")
+print(f"Classified as: {'INJECTION' if is_injection else 'SAFE'}")
+```
+
+## Limitations
+
+- Maximum input length: 2048 tokens
+- Trained on English prompts
+- Stride-based tokenization covers longer sequences with overlap
+- Use with confidence threshold tuning for your specific use case
+
+## Citation
+
+```bibtex
+@model{rafaelsandroni/modernguard-mmBERT-base-4-cross_entropy}
+```