license: apache-2.0 base_model: Qwen/Qwen2.5-Coder-3B tags: - qwen2.5 - qwen2.5-coder - coding - fine-tuned - qlora - gguf language: - id - en

⚡ ZNX-Coder-v1

ZNX-Coder-v1 is a fine-tuned coding AI built on top of Qwen2.5-Coder-3B, developed by ZNX-ai (Zenith System).

Specialized for code generation, debugging, and programming assistance in both English and Indonesian.

Model Details

Base Model	Qwen/Qwen2.5-Coder-3B
Fine-tune Method	QLoRA (r=32, alpha=64)
Epochs	3
Learning Rate	2e-5
Hardware	NVIDIA T4
Developer	ZNX-ai
Website	ZenithSystem.netlify.app

Files

File	Description
znx-coder-v1-Q4_K_M.gguf	4-bit quantized GGUF — ready to run with llama.cpp
adapter/	LoRA adapter weights

Quickstart

llama.cpp

./llama-cli -m znx-coder-v1-Q4_K_M.gguf \
    --chat-template chatml \
    -cnv

Python

from llama_cpp import Llama
from huggingface_hub import hf_hub_download

model_path = hf_hub_download(
    repo_id="RikZD/znx-coder-v1",
    filename="znx-coder-v1-Q4_K_M.gguf"
)

llm = Llama(model_path=model_path, n_ctx=2048)

response = llm.create_chat_completion(
    messages=[
        {
            "role": "system",
            "content": "You are ZNX-Coder-v1, a coding AI developed by Zenith System."
        },
        {
            "role": "user",
            "content": "Write a fibonacci function in Python"
        }
    ],
    temperature=0.1,
    max_tokens=512,
)
print(response["choices"][0]["message"]["content"])

System Prompt

You are ZNX-Coder-v1, a coding AI based on Qwen2.5-Coder-3B,
fine-tuned by Zenith System.

Limitations

• Small training dataset (121 conversations) — occasional hallucinations may occur
• Optimized primarily for Indonesian language interactions
• Focused on coding tasks, not general-purpose assistant

License

Apache 2.0 ```