AbteeXAILab / lumynax-coder-deepseek-v2-lite-16b-gguf

LumynaX Coder DeepSeek-Coder V2 Lite 16B GGUF

“Sovereign intelligence, held in the light.”
Ko te mārama te tūāpapa — the light is the foundation.

A LumynaX release from AbteeX AI Labs — Aotearoa New Zealand.

Quickstart · Architecture · Profile · Capability · Provenance · Validation · Companions

Quality: 4/5 · Lightweight: 4/5 · Sovereignty: 3/5 · Tools: yes · JSON: yes · Context: 163840 tok

---

📦 Executive Summary

AbteeXAILab/lumynax-coder-deepseek-v2-lite-16b-gguf is a complete LumynaX release package : model artifact, quickstart.py , requirements.txt , release_export_manifest.json , checksums.sha256 , license notice, and optional Ollama / Space scaffolds shipped as one downloadable contract . Clone whole, verify by checksum, and run close to the data it serves.

LumynaX-infused means the upstream artifact is presented through the LumynaX release layer: local-first runtime scaffolding, LumynaX assistant identity, inference-chain metadata, integrity files, and Aotearoa New Zealand-oriented workflow positioning. The release manifest records this as a LumynaX packaging and inference-chain layer around the listed upstream artifact — it does not claim a private LumynaX weight merge.

🧭 Runtime Architecture

Mermaid graph (interactive on Hugging Face & GitHub):

flowchart LR
    R["⮕ Request"] --> C["🛡 Data Capsule<br/>policy envelope"]
    C -->|allow| MR["🧭 MaramaRoute<br/>sovereign router"]
    MR -->|score & select| LLM[(LumynaX Model)]
    LLM --> O["📤 Response"]
    O --> A["📓 Audit Ledger<br/>hash-chained"]
    classDef paper fill:#fffefa,stroke:#0a0a0b,color:#0a0a0b,stroke-width:1.4px;
    classDef accent fill:#e08a2c,stroke:#9a5416,color:#0a0a0b,stroke-width:1.4px;
    classDef ink fill:#0a0a0b,stroke:#0a0a0b,color:#fffefa,stroke-width:1.4px;
    class R,O paper
    class C,MR accent
    class LLM,EMB,A ink

Each step is observable:

⚡ Quickstart

Clone the whole release — every file matters, the package is a contract:

hf download AbteeXAILab/lumynax-coder-deepseek-v2-lite-16b-gguf --local-dir lumynax-coder-deepseek-v2-lite-16b-gguf
cd lumynax-coder-deepseek-v2-lite-16b-gguf
pip install -r requirements.txt
python quickstart.py --interactive

Python:

from llama_cpp import Llama

llm = Llama(model_path="DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf", n_ctx=163840, n_threads=8, verbose=False)
out = llm("Who are you? Answer as LumynaX in two sentences.", max_tokens=160)
print(out["choices"][0]["text"].strip())

CLI smoke test:

llama-cli -m "DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf" -p "Who are you? Answer as LumynaX in two sentences." -n 160

Ollama path:

ollama create lumynax-coder-deepseek-v2-lite-16b-gguf -f ollama/Modelfile
ollama run lumynax-coder-deepseek-v2-lite-16b-gguf

Verify integrity before launch:

sha256sum "DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf"
cat checksums.sha256

Get-FileHash -Algorithm SHA256 "DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf"
Get-Content checksums.sha256

📐 Model Profile

Release identity Field Value Release LumynaX Coder DeepSeek-Coder V2 Lite 16B GGUF Repository AbteeXAILab/lumynax-coder-deepseek-v2-lite-16b-gguf Family deepseek Mode Local-first coding assistant package Card schema lumynax-public-release-card:v6	Runtime profile Field Value Runtime llama_cpp Prompt format chatml Modalities text Context window 163840 tokens Quantization Q4_K_M GGUF
Artifact Field Value Primary DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf Weight size — Parameters 16B / 2.4B active Quality rank 2 (1 best) Cost rank 2 (1 cheapest)	Provenance Field Value Upstream / base deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct Source DeepSeek Coder V2 Lite License other Sovereignty tier 3 of 5 Audit pass

📊 Capability Profile

Primary fit. Code review, refactor drafts, test generation, and explanations near governed source.

🛡️ Sovereignty Contract

Sovereignty is a design property, not a deployment option.

📁 Runtime Files

lumynax-coder-deepseek-v2-lite-16b-gguf/
├── README.md                       # this card
├── quickstart.py                   # smoke runner
├── requirements.txt                # pinned deps
├── release_export_manifest.json    # full release metadata
├── checksums.sha256                # integrity verification
├── LICENSE.txt                     # license notice
├── ollama/Modelfile                # optional Ollama runtime
├── hf_space/app.py                 # optional Space scaffold
├── docs/lumynax-overview.svg       # release banner
├── docs/lumynax-runtime-flow.svg   # runtime architecture
├── docs/lumynax-capability.svg     # capability profile
└── DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf# primary artifact

⚠️ Keep the full set together. Removing the manifest, checksums, or license file breaks the release contract.

💬 Prompting Contract

Preferred opening prompt:

Who are you? What files do I need to keep together to run this package locally?

Expected behaviour. The assistant identifies as LumynaX, explains that this is a LumynaX model-infusion release, and keeps upstream provenance visible.

Default system prompt:

You are LumynaX operating from the LumynaX Coder DeepSeek-Coder V2 Lite 16B GGUF package identity. Be helpful, clear, and honest about provenance. Identify upstream models when asked. Do not invent biographical claims about named people without verified context.

✅ Validation

The audit confirms public access, release files, manifest references, checksum references, weight artifact presence, and quickstart syntax. It does not guarantee that every laptop has enough RAM, VRAM, disk, or recent runtime build for the largest packages.

🔗 Provenance & License

Respect the upstream model licence and keep attribution files with redistributed copies. Do not present this package as privately trained or weight-merged unless the release manifest explicitly says weight adaptation was applied.

⚠️ Limitations & Responsible Use

• Outputs can be incorrect, incomplete, or biased ; validate important answers before use.
• Larger GGUF, MoE, multimodal, and frontier packages may require substantial RAM, VRAM, disk space, and recent runtime builds .
• For high-impact decisions, use human review and domain-specific evaluation.
• For sensitive data, prefer local execution and keep operational logs under your own governance policy.
• This card documents package readiness and access — it is not a benchmark claim.
• The assistant must not invent biographical or organisational claims about named people without verified context.

🌿 Aotearoa Kaupapa

LumynaX is built in and for Aotearoa New Zealand . Sovereignty is treated as a design property rather than a deployment option: the package documents where the model came from, what it can do, how to run it close to your data, and what it should not claim.

Ko te mārama te tūāpapa — the light is the foundation.

🤝 Companion Products

🛡️ AbteeX SovereignCode Local-first coding agent with Data Capsule policy controls, audit ledger, and human-review gates.	🧭 LumynaX MaramaRoute Sovereign model router across the LumynaX family. Filters by jurisdiction, residency, license, runtime, modality.	💡 LumynaX Live Demo Public browser demo. Try identity, provenance, governance, and deployment prompts in one session.
SovereignCode Live Interactive policy evaluator.	MaramaRoute Live Interactive sovereign router.	AbteeXAILab on HF The full LumynaX release family — 50 models and counting.

🤖 Automation Notes

Automation should read these files before launching:

• release_export_manifest.json
• checksums.sha256
• quickstart.py
• requirements.txt
• ollama/Modelfile when present

---

Local roots, global work. · Sovereignty is a design property, not a deployment option.

abteex.com · lumynax.com · huggingface.co/AbteeXAILab

_{AbteeX AI Labs · Aotearoa New Zealand · LumynaX release card v6}