license: apache-2.0
base_model: google/gemma-4-12b
language:
- en
- multilingual
tags:
- code
- coding
- agentic
- gemma-4
- code-generation
- code-review
- debugging
- instruction-tuned
- perciqa
- aurora
- canadian-ai
- mini
- lora
- sft
datasets:
- perciqa/aurora-code-sft-v1
pipeline_tag: text-generation
library_name: transformers
Aurora-Code-Mini-1
Compact. Capable. Completely open.
Aurora-Code-Mini-1 is a 12B dense coding model built by Perciqa, a Canadian AI company. Fine-tuned from Gemma 4 12B on curated agentic coding instruction pairs, Aurora-Code-Mini-1 is designed for developers who need fast, high-quality coding assistance on consumer hardware — without cloud dependencies, usage limits, or black boxes.
3B active parameters. Apache 2.0 license. Fits on a 24 GB GPU at BF16 or 8 GB cards with quantization.
What Aurora-Code-Mini-1 does
Aurora-Code-Mini-1 brings the Aurora coding experience to hardware that fits on your desk. Run it locally, integrate it into your toolchain, and keep your code private.
- Code generation — write functions, classes, and complete programs across 40+ languages
- Debugging — identify root causes and produce clear, actionable fixes
- Code review — flag issues, suggest refactors, explain tradeoffs
- Agentic tasks — multi-step reasoning, tool use, and repository-level workflows
No cloud required. No data leaving your machine. Your model, your terms.
Quickstart
Install
pip install "transformers>=4.51.0" accelerate
Hardware:
24 GB VRAM at BF16. Works on a single RTX 3090/4090, A5000, or equivalent. For smaller setups, use 4-bit quantization (8 GB VRAM).
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "Perciqa/Aurora-Code-Mini-1"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto",
)
system_prompt = (
"You are Aurora, an AI code assistant built by Perciqa. "
"You help developers write, review, and understand code. "
"You provide clear, correct, and complete solutions. "
"When you're unsure, you say so."
)
messages = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": "Write a Python function to merge two sorted lists."},
]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([text], return_tensors="pt").to(model.device)
outputs = model.generate(
**inputs,
max_new_tokens=1024,
temperature=0.7,
do_sample=True,
)
response = tokenizer.decode(outputs[0][len(inputs.input_ids[0]):], skip_special_tokens=True)
print(response)
4-bit Quantization (8 GB GPU)
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
quantization_config = BitsAndBytesConfig(load_in_4bit=True)
model = AutoModelForCausalLM.from_pretrained(
"Perciqa/Aurora-Code-Mini-1",
quantization_config=quantization_config,
device_map="auto",
)
vLLM (recommended for production)
pip install vllm
vllm serve Perciqa/Aurora-Code-Mini-1 --max-model-len 32768
Query via the OpenAI-compatible API:
from openai import OpenAI
client = OpenAI(base_url="http://localhost:8000/v1", api_key="token-abc123")
response = client.chat.completions.create(
model="Perciqa/Aurora-Code-Mini-1",
messages=[
{"role": "system", "content": "You are Aurora, an AI code assistant built by Perciqa."},
{"role": "user", "content": "Refactor this function to be more Pythonic."},
],
max_tokens=1024,
)
print(response.choices[0].message.content)
Ollama
ollama run hf.co/Perciqa/Aurora-Code-Mini-1
Argus Integration
Argus is Perciqa's open-source agent observability SDK — trajectory tracing, token usage, eval-in-production. Aurora-Code-Mini-1 integrates natively.
import argus
from openai import OpenAI
argus.init(server_url="http://localhost:8000", agent_name="aurora-mini")
client = OpenAI(base_url="http://localhost:8000/v1", api_key="token-abc123")
@argus.trace(kind="agent")
def aurora_mini_code(query: str) -> str:
response = client.chat.completions.create(
model="Perciqa/Aurora-Code-Mini-1",
messages=[
{"role": "system", "content": "You are Aurora, an AI code assistant built by Perciqa."},
{"role": "user", "content": query},
],
)
return response.choices[0].message.content
result = aurora_mini_code("Write a TypeScript function that validates an email address.")
print(result)
Argus captures latency, token usage, inputs/outputs, and agent spans — visible in the Argus dashboard with no extra instrumentation.
Performance
Benchmarks in progress. Independent evaluations on LiveCodeBench, HumanEval+, and MBPP+ will be published here before the stable release.
Model Details
| Field | Value |
|---|---|
| Architecture | Dense Transformer |
| Total Parameters | 12B |
| Base Model | Gemma 4 12B |
| Fine-Tuning | LoRA SFT — curated agentic coding pairs |
| Context Length | 128,000 tokens |
| License | Apache 2.0 |
| Hardware (BF16) | 24 GB VRAM |
| Hardware (4-bit) | ~8 GB VRAM |
Training
Aurora-Code-Mini-1 is trained with LoRA supervised fine-tuning on a curated dataset of coding instruction pairs. Categories covered:
- Code generation (Python, TypeScript, Go, Rust, and more)
- Debugging and root cause analysis
- Code review and refactoring
- Multi-step agentic reasoning and tool use
Roadmap
| Version | Description | Status |
|---|---|---|
| v1 | LoRA SFT on curated agentic coding pairs. Perciqa system prompt and Argus integration. | Released |
| v2 | Expanded SFT dataset. Independent benchmark evaluation. | Q3 2026 |
| v3 | Full fine-tune with extended dataset across generation, debugging, review, and test writing. | Q4 2026 |
System Prompt
You are Aurora, an AI code assistant built by Perciqa. You help developers write, review, and understand code. You provide clear, correct, and complete solutions. When you're unsure, you say so.
About Perciqa
Perciqa is a Canadian AI company building enterprise models and tools that organisations can deploy, audit, and fully control — on their own infrastructure, on their own terms. Founded in 2023 and based in Canada 🇨🇦.
License
Aurora-Code-Mini-1 is released under the Apache 2.0 License.
Made with ♥ by Perciqa 🇨🇦