spst01/SmolLM2-135M-Sukshma

⚡ SmolLM2-135M-Instruct (Sūkshma-LLM Ultra-Compact Edition)

Engineered by Sovereign Byte Technology

📩 Contact: sovereignbyte.tech@gmail.com

---

🌟 Overview

SmolLM2-135M-Instruct (.sukshma) is an ultra-compact edge AI artifact engineered by Sovereign Byte Technology. Built for mobile phones, IoT gateways, microcontrollers, and low-power Apple Silicon hardware, this model packs 134.5 Million parameters into just 63.42 MB—achieving a 4.05x physical storage reduction (75.29% space saved) from original uncompressed footprints with 90.69% mean layer cosine fidelity.

Powered by Sovereign Byte Technology's Sovereign Sub-Byte Discrete Compression (SSDC™) and Adaptive Salient Feature Preservation, Sūkshma eliminates compute bottlenecks on resource-constrained devices, dramatically lowering thermal load and memory bandwidth requirements.

---

📊 Benchmark: SafeTensors vs .sukshma (Apple Silicon M1)

Benchmark Metric	Official SafeTensors	Sūkshma (.sukshma)	Sovereign Advantage
Physical Disk Footprint	256.60 MB	63.42 MB	4.05x Smaller (75.29% Space Saved)
Active Mobile RAM	~1.1 GB resident	~245 MB resident	Runs on Raspberry Pi & MicroVMs
Mean Layer Cosine Fidelity	100.0% (FP32)	90.69%	High semantic fidelity preservation
Matrix Projection Latency	30.17 µs	10.97 µs	2.75x FASTER Execution
Edge Throughput	33,145 layer-ops/s	91,137 layer-ops/s	Hardware-Optimized ALU Execution
Energy Consumption	Standard FMA Load	Low-Power Integer Execution	~60% Lower Battery Drain

---

🔌 Drop-In API Server (LM Studio, Ollama, Open-WebUI & Cursor Compatible)

Run .sukshma models with any local AI desktop tool, IDE, or client library with zero dependencies:

# Clone the repository
git clone https://huggingface.co/spst01/SmolLM2-135M-Sukshma
cd SmolLM2-135M-Sukshma

# Launch local OpenAI & Ollama compatible server
python3 sukshma_runner.py serve SmolLM2-135M-Instruct.sukshma --port 11434

Instant Client Compatibility:

• LM Studio: Under "Connect Custom Server", set endpoint to http://localhost:11434/v1.
• Open-WebUI / AnythingLLM: Set OPENAI_API_BASE="http://localhost:11434/v1".
• Cursor / VS Code: Set OpenAI Base URL to http://localhost:11434/v1.
• cURL Command:

  curl http://localhost:11434/v1/chat/completions \
    -H "Content-Type: application/json" \
    -d '{"messages": [{"role": "user", "content": "Explain edge AI in one sentence"}]}'
  

---

🚀 Standalone Python Runner

# 1. Inspect container metadata & tensor breakdown
python3 sukshma_runner.py inspect SmolLM2-135M-Instruct.sukshma

# 2. Run instant layer-level forward projection benchmark
python3 sukshma_runner.py benchmark SmolLM2-135M-Instruct.sukshma

# 3. Verify bit-exact tensor reconstruction
python3 sukshma_runner.py verify SmolLM2-135M-Instruct.sukshma

---

🌐 In-Browser WebAssembly (WASM) Execution

Open browser_demo.html directly in any modern browser (Chrome, Safari, Firefox, Edge, Mobile Safari) to run client-side execution with zero server dependencies:

# Launch a local server or open browser_demo.html directly
open browser_demo.html

Or embed via WebAssembly in your frontend:

// Parse .sukshma binary array buffer directly in the browser
const buffer = await fetch("SmolLM2-135M-Instruct.sukshma").then(r => r.arrayBuffer());
const view = new DataView(buffer);
const magic = view.getUint32(0, true); // 0x53554B53 ("SUKS")
console.log("Sūkshma container verified client-side!");

---

🏢 About Sovereign Byte Technology

Sovereign Byte Technology is an enterprise deep-tech research lab developing sovereign, ultra-low-latency AI engines and ultra-compact edge execution containers.

• 📩 Commercial & Business Inquiries: sovereignbyte.tech@gmail.com

For commercial licensing, enterprise deployment, or custom model quantization, please email us directly at sovereignbyte.tech@gmail.com.

---

📜 Citation & License

@software{sovereignbyte_sukshma_smollm2_2026,
  author = {Sovereign Byte Technology},
  title = {Sūkshma-LLM: Ultra-Compact Edge AI Container Engine},
  year = {2026},
  publisher = {Hugging Face},
  journal = {Hugging Face Model Hub},
  howpublished = {\url{https://huggingface.co/spst01/SmolLM2-135M-Sukshma}}
}

Base weights licensed under Apache 2.0 by Hugging Face TB. Sūkshma containerization and edge tooling © 2026 Sovereign Byte Technology.