README.md

Sentinel-D DistilBERT Intent Classifier v1

Model Details

• Base Model: distilbert-base-uncased
• Task: Sequence Classification (4 classes)
• Training Date: 2026-03-04T03:36:30.927788
• Classes: VERSION_PIN, API_MIGRATION, MONKEY_PATCH, FULL_REFACTOR

Intent Classes

1. VERSION_PIN: Pinning or locking dependency versions
2. API_MIGRATION: Migrating between API versions or protocols
3. MONKEY_PATCH: Quick/temporary fixes via runtime patching
4. FULL_REFACTOR: Complete rewriting or structural redesign

Usage

from transformers import DistilBertTokenizer, DistilBertForSequenceClassification
import torch
import json

tokenizer = DistilBertTokenizer.from_pretrained("./distilbert-intent-classifier-v1")
model = DistilBertForSequenceClassification.from_pretrained("./distilbert-intent-classifier-v1")

text = "I updated my package.json to lock the Express version to 4.18.0"
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)

with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits
    predicted_class_id = logits.argmax(-1).item()

# Map ID back to label
label_config = json.load(open("label_config.json"))
predicted_label = label_config["id_to_label"][str(predicted_class_id)]
print(f"Predicted Intent: {predicted_label}")

Files

• pytorch_model.bin: Fine-tuned model weights
• config.json: Model configuration
• vocab.txt: Tokenizer vocabulary
• label_config.json: Intent class mappings
• README.md: This file

Training Configuration

• Epochs: 6
• Batch Size: 16
• Learning Rate: Dynamic tuning from 1e-05 to 5e-05
• Optimizer: AdamW with weighted cross-entropy loss
• Class Imbalance Handling: Random oversampling + weighted loss

Performance Targets

• Accuracy: >= 0.80
• Macro F1: >= 0.80