Siddh07ETH/Atlas-Coder-2-0.5B-GGUF

⚡ Atlas-Coder-2-0.5B GGUF

Quantized GGUF releases of Atlas-Coder-2-0.5B, optimized for llama.cpp, Ollama, and LM Studio.

---

🏆 EvalPlus Strict Benchmarks

Atlas-Coder-2 is a 0.5B parameter coding model fine-tuned on 50,000 execution-verified Python programming samples.

Despite its compact size, it achieves state-of-the-art performance among strictly sub-1B coding models on the EvalPlus benchmark suite.

Benchmark	Atlas-Coder-2	Qwen2.5-Coder-0.5B	DeepSeek-Coder-1.3B	Llama-3.2-1B
HumanEval+	36.6% 🥇	34.1%	35.4%	12.2%
MBPP+	43.9% 🥇	42.1%	39.8%	25.6%

Evaluation: EvalPlus (Pass@1, Greedy Decoding)

---

📦 Available Quantizations

File	Quantization	Size	Recommended For
Atlas-Coder-2-0.5B-F16.gguf	F16	948 MB	Maximum accuracy & benchmarking
Atlas-Coder-2-0.5B-Q8_0.gguf	Q8_0	506 MB	Near-lossless inference
Atlas-Coder-2-0.5B-Q6_K.gguf	Q6_K	482 MB	Best quality / size balance
Atlas-Coder-2-0.5B-Q5_K_M.gguf	Q5_K_M	401 MB	Recommended for most users ⭐
Atlas-Coder-2-0.5B-Q4_K_M.gguf	Q4_K_M	379 MB	Low-memory devices

---

🚀 Quick Start

Ollama

Create a file named Modelfile

FROM ./Atlas-Coder-2-0.5B-Q5_K_M.gguf

SYSTEM "You are Atlas-Coder, an elite AI coding assistant. You write clean, efficient, and well-documented Python code."

PARAMETER temperature 0.2
PARAMETER top_p 0.95
PARAMETER repeat_penalty 1.1

PARAMETER stop "<|im_start|>"
PARAMETER stop "<|im_end|>"

Create the model:

ollama create atlas-coder-2 -f Modelfile

Run:

ollama run atlas-coder-2 "Write a Python function to check whether a string is a palindrome."

---

LM Studio

1. Open LM Studio
2. Search for Siddh07ETH/Atlas-Coder-2-0.5B-GGUF
3. Download Q5_K_M or Q6_K
4. Load the model and start coding.

---

llama.cpp

./llama-cli \
  -m Atlas-Coder-2-0.5B-Q5_K_M.gguf \
  -p "<|im_start|>system\nYou are a helpful coding assistant.<|im_end|>\n<|im_start|>user\nWrite a Python binary search implementation.<|im_end|>\n<|im_start|>assistant\n" \
  -n 512 \
  --temp 0.2 \
  --top-p 0.95 \
  --repeat-penalty 1.1

---

💬 Chat Template

Atlas-Coder-2 uses the standard ChatML prompt format.

<|im_start|>system
{system_message}
<|im_end|>

<|im_start|>user
{user_message}
<|im_end|>

<|im_start|>assistant

---

📚 Open-Source Ecosystem

Base Model

• Siddh07ETH/Atlas-Coder-2-0.5B

Training Dataset

• Siddh07ETH/Atlas-Coder-50K-ChatML
• 50,000 execution-verified Python instruction samples in ChatML format.

---

👨‍💻 Author

Siddharth N.R.
Pluto AI Research

---

📄 License

This repository is released under the Apache 2.0 License.

The base model (Qwen2.5-Coder-0.5B-Instruct) is also licensed under Apache 2.0.