| --- |
| language: en |
| license: mit |
| tags: |
| - text-classification |
| - gpt2 |
| - transformers |
| - pytorch |
| - custom-architecture |
| - tiktoken |
| library_name: transformers |
| --- |
| |
| # TicketClassificationGPT |
|
|
| ## Model Summary |
|
|
| **TicketClassificationGPT** is a GPT-2–based text classification model designed entirely from scratct to classify IT support tickets into 8 predefined categories. |
| The model uses the original OpenAI GPT-2 architecture and weights, with the language modeling head replaced by a custom classification head. Only the final layers were fine-tuned for the ticket classification task. |
|
|
| This model is fully compatible with the Hugging Face `transformers` ecosystem and can be loaded using `AutoModel.from_pretrained`. |
|
|
| --- |
|
|
| ## How to Get Started with the Model |
|
|
| ### Inference Example (Transformers + tiktoken) |
|
|
| ```python |
| from transformers import AutoModel |
| import tiktoken |
| |
| # Load tokenizer |
| tokenizer = tiktoken.get_encoding("gpt2") |
| |
| # Load model |
| model_id = "FarhanAK128/TicketClassificationGPT" |
| model = AutoModel.from_pretrained( |
| model_id, |
| trust_remote_code=True |
| ) |
| |
| # Example prediction |
| text = "Need extra space on Google Drive." |
| prediction = model.predict(text, tokenizer) |
| |
| print("Predicted class:", prediction) # Predicted class: Storage |
| ``` |
|
|
| **Note:** This model uses a custom `.predict()` method defined in the repository and requires `trust_remote_code=True` to function. |
|
|
| --- |
|
|
| ## Model Details |
|
|
| ### 📝 Model Description |
|
|
| - **Developed by:** Farhan Ali Khan |
| - **Model type:** GPT-2–like text classification model |
| - **Framework:** PyTorch |
| - **Task:** Text Classification |
| - **Number of classes:** 8 |
| - **Language:** English |
| - **License:** MIT |
|
|
| ### 📋 Classification Labels |
|
|
| | Class ID | Category | |
| |----------|--------------------------| |
| | 0 | Hardware | |
| | 1 | HR Support | |
| | 2 | Access | |
| | 3 | Miscellaneous | |
| | 4 | Storage | |
| | 5 | Purchase | |
| | 6 | Internal Project | |
| | 7 | Administrative Rights | |
|
|
|
|
| ### Model Sources |
|
|
| - **Repository:** https://huggingface.co/FarhanAK128/TicketClassificationGPT |
| - **Base model:** OpenAI GPT-2 like architecture from scratch |
| - **Paper:** https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf |
| |
| --- |
| |
| ## Training Details |
| ### Training Data |
| |
| The model was trained on the **IT Service Ticket Classification Dataset** available on Kaggle. |
| |
| - **Dataset name:** IT Service Ticket Classification Dataset |
| - **Source:** Kaggle |
| - **Link:** https://www.kaggle.com/datasets/adisongoh/it-service-ticket-classification-dataset |
| - **Content:** Labeled IT support ticket text data |
| - **Language:** English |
| |
| The dataset was used for supervised multi-class classification after standard text preprocessing and tokenization. |
| |
| |
| ### Training Procedure |
| |
| - **Base weights:** OpenAI GPT-2 |
| - **Fine-tuning strategy:** Partial fine-tuning (classification head + final transformer layers) |
| - **Optimizer:** AdamW |
| - **Learning rate:** 1e-4 |
| - **Weight decay:** 0.1 |
| - **Epochs:** 5 |
| - **Random seed:** 123 |
| - **Loss function:** Cross-Entropy Loss |
| - **Training regime:** FP32 |
| - **Evaluation frequency:** Every 30 steps |
| - **Total training time:** ~140 minutes |
| - **Final training loss:** ~0.61 |
| - **Final validation loss:** ~0.86 |
| |
| ### 📈 Training Progress |
| #### Training and Validation Loss |
|  |
| |
| |
| #### Training and Validation Accuracy |
|  |
| |
| ### 📊 Model Performance |
| |
| | Dataset Split | Accuracy | |
| |--------------|----------| |
| | 🏋️ Training | **76.54%** | |
| | 🧪 Validation | **75.67%** | |
| | 🧠 Test | **73.83%** | |
| |
| --- |
| |
| ## Uses |
| |
| ### Direct Use |
| |
| This model can be used directly to classify short IT support ticket texts into predefined categories. |
| |
| Example use cases: |
| - Automated ticket routing |
| - Helpdesk prioritization |
| - Internal IT workflow automation |
| |
| ### Downstream Use |
| |
| The model may be further fine-tuned on: |
| - Organization-specific ticket data |
| - Expanded label sets |
| - Domain-specific terminology |
| |
| ### Out-of-Scope Use |
| |
| - Multilingual text classification |
| - Open-domain topic classification |
| - Legal, medical, or safety-critical decision-making |
| |
| --- |
| |
| ## Bias, Risks, and Limitations |
| |
| - Trained on a limited-domain dataset (IT support tickets) |
| - Not evaluated for demographic or social bias |
| - Predictions may be unreliable for unseen ticket categories |
| - Performance depends on input text quality and length |
| |
| ### Recommendations |
| |
| Human validation is recommended before using predictions in production systems. |
| For best results, further fine-tuning on in-domain data is advised. |
| |
| --- |
| |
| ## Model Card Authors |
| **Farhan Ali Khan** |
| |
| ## Model Card Contact |
| For questions or feedback, please reach out via my Hugging Face profile: |
| [FarhanAK128](https://huggingface.co/FarhanAK128) |
