Update README.md

README.md

@@ -1,3 +1,89 @@
----
-license: mit
----
+---
+license: mit
+datasets:
+- RnniaSnow/st-code-dataset
+language:
+- en
+base_model:
+- RnniaSnow/ST-Coder-14B
+tags:
+- code
+- st
+- plc
+- industry
+- gguf
+- llama.cpp
+---
+
+# ST-Coder-14B (GGUF)
+
+This repository contains the **GGUF quantized versions** of [RnniaSnow/ST-Coder-14B](https://huggingface.co/RnniaSnow/ST-Coder-14B). 
+
+**ST-Coder-14B** is an industrial-grade Large Language Model optimized for **Programmable Logic Controller (PLC)** programming, specifically focused on the **IEC 61131-3 Structured Text (ST)** language. 
+
+By providing GGUF formats, automation engineers can run this model **locally and entirely offline** on standard laptops or IPCs (Industrial PCs) using CPU or edge GPUs, which is crucial for secure and air-gapped shop-floor environments.
+
+## 💾 Available Quantization Formats
+
+We provide several quantization levels. Choose the one that best fits your hardware (RAM/VRAM):
+
+| File Name | Bit | Size | RAM Required | Recommended For |
+| :--- | :---: | :---: | :---: | :--- |
+| `st-coder-14b-q4_k_m.gguf` | 4-bit | ~8.5 GB | 12 GB+ | **Recommended.** Best balance of speed, size, and code logic preservation. |
+| `st-coder-14b-q6_k.gguf`   | 6-bit | ~11.5 GB | 16 GB+ | Very high quality, minimal precision loss. |
+| `st-coder-14b-q8_0.gguf`   | 8-bit | ~15.2 GB | 20 GB+ | Near-lossless precision for complex engineering math/logic. |
+
+> *Note: Code generation models are sensitive to heavy quantization. We do not recommend using anything below Q4 for ST code generation, as syntax accuracy may degrade.*
+
+## 🚀 How to Use
+
+Since this model is based on the Qwen2.5 architecture, it uses the **ChatML** prompt format. Most modern tools will auto-detect this from the GGUF metadata.
+
+### Method 1: LM Studio (Easiest GUI)
+This is the recommended method for Windows/macOS users who prefer a graphical interface.
+1. Download and install [LM Studio](https://lmstudio.ai/).
+2. Search for `RnniaSnow/ST-Coder-14B-GGUF` in the top search bar.
+3. Download the `q4_k_m` or `q6_k` file.
+4. Load the model, set the system prompt to: *"You are an expert industrial automation engineer specializing in IEC 61131-3 Structured Text."* and start chatting.
+
+### Method 2: Ollama (CLI & API)
+You can easily serve this model using [Ollama](https://ollama.com/).
+1. Download your preferred `.gguf` file to your local machine.
+2. Create a file named `Modelfile` in the same directory with the following content:
+```dockerfile
+FROM ./st-coder-14b-q4_k_m.gguf
+TEMPLATE """<|im_start|>system
+{{ .System }}<|im_end|>
+<|im_start|>user
+{{ .Prompt }}<|im_end|>
+<|im_start|>assistant
+"""
+SYSTEM """You are an expert industrial automation engineer specializing in IEC 61131-3 Structured Text."""
+PARAMETER temperature 0.2
+PARAMETER top_p 0.9
+
+```
+
+3. Create and run the model in your terminal:
+
+```bash
+ollama create st-coder -f Modelfile
+ollama run st-coder "Write a Function Block for a PID controller."
+
+```
+
+### Method 3: llama.cpp (Command Line)
+
+For advanced users deploying via `llama.cpp`:
+
+```bash
+./llama-cli -m st-coder-14b-q4_k_m.gguf -p "<|im_start|>system\nYou are an expert PLC programmer.<|im_end|>\n<|im_start|>user\nWrite an ST program for a conveyor belt motor.<|im_end|>\n<|im_start|>assistant\n" -n 1024 -c 8192 --temp 0.2
+
+```
+
+## ⚠️ Disclaimer & Industrial Safety
+
+**Industrial Control Systems (ICS) carry significant physical risks.** * This AI model generates code based on statistical probabilities and does **not** guarantee logical correctness, real-time safety, or hardware compatibility.
+
+* **Always** verify, simulate, and strictly test the generated code in a safe environment before deploying it to physical hardware (PLCs, drives, robotics).
+* The creators of this model assume absolutely no liability for any damage, injury, or production downtime resulting from the use of this code.