README.md

---
license: apache-2.0
base_model: Qwen/Qwen2.5-Coder-14B-Instruct
---

# 🔌 AutonomusHDL — Verilog-Finetuned Qwen2.5-Coder-14B (GGUF)

**AutonomusHDL** is a fine-tuned version of [Qwen2.5-Coder-14B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-14B-Instruct) specifically optimized for **Hardware Description Language (HDL)** tasks, with a focus on **Verilog** code generation, completion, and reasoning. The model is provided in GGUF format for efficient local inference via `llama.cpp` and compatible runtimes.

---

## 📦 Available Files

| File | Quantization | Size | Use Case |
|---|---|---|---|
| `qwen2.5_coder_14b_instruct_verilog_finetuned_q8.gguf` | Q8_0 | 15.7 GB | Highest quality, more VRAM/RAM |
| `Qwen2.5 coder-14B-Q3_K_L.gguf` | Q3_K_L | 7.9 GB | Lighter, faster, lower memory footprint |

> **Recommendation:** Use the **Q8** model if you have ≥16 GB RAM/VRAM for best output quality. Use **Q3_K_L** for systems with limited resources.

---

## 🧠 Model Details

| Property | Value |
|---|---|
| **Base Model** | Qwen2.5-Coder-14B-Instruct |
| **Fine-tune Domain** | Verilog / HDL Code Generation |
| **Format** | GGUF |
| **License** | Apache 2.0 |
| **Parameters** | 14B |
| **Context Length** | Up to 128K tokens (base model) |

---

## 🚀 Quickstart

### With `llama.cpp`

```bash
# Clone and build llama.cpp
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp && make

# Run inference
./llama-cli \
  -m qwen2.5_coder_14b_instruct_verilog_finetuned_q8.gguf \
  -p "Write a Verilog module for a 4-bit synchronous counter with reset." \
  -n 512 \
  --temp 0.2
```

### With Ollama

```bash
# Create a Modelfile
echo 'FROM ./qwen2.5_coder_14b_instruct_verilog_finetuned_q8.gguf' > Modelfile

# Import and run
ollama create autonomusHDL -f Modelfile
ollama run autonomusHDL
```

### With LM Studio

1. Download one of the `.gguf` files above.
2. Open **LM Studio** → Load Model → select the downloaded file.
3. Start chatting with Verilog prompts directly.

---

## 💡 Example Prompts

**Module generation:**
```
Write a Verilog module for a parameterized FIFO with configurable depth and width.
```

**Debugging:**
```
The following Verilog code has a timing issue. Identify and fix it:
[paste your code]
```

**Testbench generation:**
```
Generate a SystemVerilog testbench for a 32-bit ALU module with add, sub, AND, OR, and XOR operations.
```

**FSM design:**
```
Implement a Moore FSM in Verilog for a traffic light controller with states: RED, GREEN, YELLOW.
```

---

## 🎯 Intended Use Cases

- RTL design and Verilog code generation
- HDL code completion and auto-suggestions
- Testbench and assertion generation
- Debugging and explaining existing Verilog/VHDL code
- Learning and educational HDL workflows
- Integration into EDA tool pipelines

---

## ⚙️ Hardware Requirements

| Quantization | Min RAM/VRAM | Recommended |
|---|---|---|
| Q8_0 (15.7 GB) | 16 GB | 24 GB+ |
| Q3_K_L (7.9 GB) | 8 GB | 12 GB+ |

For CPU-only inference, ensure you have sufficient system RAM. GPU offloading via `llama.cpp` is supported with CUDA/Metal/Vulkan.

---

## 📜 License

This model is released under the **Apache 2.0 License**. The base model weights are subject to the [Qwen2.5 license](https://huggingface.co/Qwen/Qwen2.5-Coder-14B-Instruct/blob/main/LICENSE).

---

## 🙏 Acknowledgements

- Base model: [Qwen2.5-Coder-14B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-14B-Instruct) by Alibaba Cloud
- GGUF conversion tooling: [llama.cpp](https://github.com/ggerganov/llama.cpp) by Georgi Gerganov

---

## ✉️ Contact

For questions, issues, or collaboration, reach out via the [Community tab](https://huggingface.co/Vishvjit2001/autonomusHDL/discussions) on this repository.