---
base_model: Qwen/Qwen2.5-Coder-1.5B
tags:
- infrastructure-as-code
- terraform
- ansible
- kubernetes
- docker
- devops
- code-generation
- lora
- sft
- trl
license: apache-2.0
datasets:
- bigcode/commitpackft
language:
- en
pipeline_tag: text-generation
library_name: peft
---

# IaC-Coder 1.5B — Infrastructure as Code Language Model

A small language model fine-tuned for DevOps Infrastructure as Code (IaC) tasks. Built on [Qwen2.5-Coder-1.5B](https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B) with LoRA SFT on real-world IaC commits.

## Supported IaC Languages

| Language | Use Case | Training Data |
|----------|----------|---------------|
| **Terraform HCL** | Cloud infrastructure provisioning | 421 commit pairs |
| **YAML (Ansible/K8s)** | Configuration management, container orchestration | 30,000 commit pairs (capped from 114K) |
| **Dockerfile** | Container image definitions | 39 commit pairs |
| **Shell/Bash** | CI/CD pipelines, automation scripts | 15,000 commit pairs (capped from 31K) |
| **Nix** | Reproducible builds | 1,593 commit pairs |
| **Makefile** | Build automation | 960 commit pairs |
| **SaltStack** | Configuration management | 617 commit pairs |
| **TOML** | Application configuration | 3,424 commit pairs |

## Quick Start

```python
from transformers import pipeline
import torch

pipe = pipeline(
    "text-generation",
    model="Tejas86/iac-coder-1.5b",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are an expert DevOps engineer. Generate clean, production-ready Terraform HCL code."},
    {"role": "user", "content": "Task: Create an AWS S3 bucket with versioning enabled and server-side encryption\n\nGenerate the Terraform HCL code."},
]

output = pipe(messages, max_new_tokens=512, temperature=0.2, do_sample=True, top_p=0.9)
print(output[0]["generated_text"][-1]["content"])
```

## Prompt Format

The model was trained with this conversational format:

```
System: You are an expert DevOps engineer specializing in Infrastructure as Code. Generate clean, production-ready {language} code.

User: Task: {natural language description}

Modify the following {language} file:
```
{existing code}
```