| # Regex-Helper (Powered by ML-Forge) | |
| **Precision Regular Expression Assistant** built using a specialized fine-tuning pipeline. | |
| ## 🏗 ML-Forge Workflow | |
| This model is generated using the **ML-Forge** engine, a parameterized automation stack for rapid LLM development. | |
| ### 🚀 Rapid Start | |
| Follow these steps to go from zero to a published model: | |
| #### 1. Initialize | |
| Sets up the base Llama 3.2 weight. | |
| ```bash | |
| ./scripts/setup.sh | |
| ``` | |
| #### 2. Prepare Data | |
| Pulls `bndis/regex_instructions` from Hugging Face and cleans it. | |
| ```bash | |
| source config.sh | |
| uv run python scripts/data_prep.py | |
| ``` | |
| #### 3. Train | |
| Starts the LoRA training session (1000 iterations, Rank 16). | |
| ```bash | |
| ./scripts/train.sh | |
| ``` | |
| #### 4. Publish | |
| Fuses weights, creates GGUFs, and pushes to HF, Ollama, and Kaggle. | |
| ```bash | |
| ./scripts/publish.sh | |
| ``` | |
| ## 📊 Technical Configuration | |
| Parameters are managed in `config.sh`: | |
| - **Base**: Llama 3.2 3B Instruct | |
| - **Rank**: 16 | |
| - **Context**: 2048 tokens | |
| - **Precision**: Q4_K_M (Ollama) / BF16 (HF) | |
| --- | |
| _Created by the ML-Forge Pipeline._ | |