SKT-NRS/SKT-OMNI-CORPUS-146T-V1

ON BULIDING STAGE

🪐 SKT-OMNI-CORPUS-146T-V1

TOTAL SIZE - 5TB

VERSION - 1 0.5 To O.6 TRILLON TOKENS

TARGET - 8TB

*VERSION - 2 10 TRILLON *

The Sovereign Data Foundation for Indigenous LLM Development (Project Surya)

---

🔱 SKT AI LABS: PROJECT SURYA

---

🚀 Welcome to Project Surya (सूर्य)

Haters called it a meme. We called it a foundation. Project Surya is not just another LLM; it is India's first truly sovereign AI foundation model, trained on 146 Trillion tokens of unique Hindi and Hinglish data. While others wait for translated data, we process the native thoughts of 1.4 billion people.

🔱 About

At SKT AI LABS, we don't announce; we deploy. We don't argue; we compile. Our commitment is to building an AI that understands the logic, culture, and slang of India from Sidhi to Siliguri.

---

📌 Executive Overview

SKT-OMNI-CORPUS-146T is a premier, high-fidelity dataset architecture engineered at SKT AI LABS. It serves as the primary pre-training backbone for the ST-X Series and Project Surya.

Unlike generic crawls, this corpus is a meticulously curated mix of 5000+ high-quality data streams, sharded into a standardized 3.5GB JSONL format (SKT-𝕻 series) to ensure maximum throughput during distributed training and Mixture-of-Experts (MoE) compatibility.

---

🔥 Current Dataset Progress (SKT OMNI CORPUS V1)

Aapka ongoing grind directly reflects here. While you are processing the 120GB+ Hinglish data, the scale is building up.

Metric	Status / Progress	Impact
Hinglish Shards	SKT-𝕻-1 to SKT-𝕻-500+	Massive Logic Base
Token Density	High (Educational + Conversational)	Reasoning & Chat Flow
Identity Branding	Locked in metadata	Hater Protection

---

🔱 The "No-Skip" Policy

We process everything—from the deepest mathematical reasoning Of Aryabhatt the daily conversations of a digital India. Our pipeline ensures every token counts.

---

🏗️ Dataset Architecture & Schema

We employ a Zero-Junk Policy. Every entry undergoes rigorous cleaning, deduplication, and is tagged with permanent metadata to maintain lineage and developer identity.

🛡️ Data Schema (JSONL)

{
  "text": "High-quality curated educational or linguistic content...",
  "skt_id": "uuid-v4-identifier",
  "Author": "𝙎𝙆𝙏 𝘼𝙄 𝙇𝘼𝘽𝙎",
  "origin": "Sidhi, MP, India",
  "timestamp": "2026-03-23T21:18:39Z"
}

Join the Innovation

Bhai, stay tuned for the training logs. Project Surya 1.1T MoE is coming to take over.

🕉️ LOVE ❤	TO INNOVATAVE 🕉️

---