Skip to content

README.md

Latest commit

 

History

History
190 lines (135 loc) · 7.62 KB

README.md

File metadata and controls

190 lines (135 loc) · 7.62 KB
Raw

SiMaSeVaKuGa LLM

Sovereign Tamil Language Model

Silappatikaram · Manimekalai · Seevaga Sinthamani · Valayapathi · Kundalakesi · Ga (Muruga — தமிழின் தோற்றம்)

License: Apache 2.0 Tamil Status GitHub Stars

What is SiMaSeVaKuGa LLM?

SiMaSeVaKuGa is a homegrown, sovereign Large Language Model (LLM) — a vLLM/sLLM platform delivering cutting-edge language model capabilities tailored specifically for the Tamil language and Indian linguistic contexts.

A robust model capable of understanding, generating, and reasoning in Tamil — built by Tamil people, for Tamil people.

Named after the Five Great Tamil Epics (ஐம்பெரும் காப்பியங்கள்) and Ga honouring Muruga — deeply intertwined with the origin of Tamil — this model is trained on the richest corpus of Tamil literary heritage ever assembled for machine learning.

Why SiMaSeVaKuGa?

Tamil is one of the world's oldest living classical languages. Yet Tamil speakers — 80+ million people — are underserved by English-centric global LLMs that fail to capture Tamil's grammatical depth, dialectal richness, and cultural nuance.

SiMaSeVaKuGa LLM changes that:

Problem SiMaSeVaKuGa Solution
Global LLMs hallucinate in Tamil Trained natively on Tamil corpus
No Tamil-first GenAI tooling Open API for Tamil GenAI & Agentic AI
English-centric reasoning Tamil-optimized reasoning & generation
No Tamil voice AI Dialect-aware voice inference
Tamil culture lost in translation Trained on classical Tamil literature

Training Corpus

The model is trained on Tamil's documented literary and linguistic heritage:

Source Description Est. Date
Tolkappiam (தொல்காப்பியம்) World's oldest surviving Tamil grammar — 3 chapters: Ezhuthu, Soll, Porul 500 BCE – 300 CE
Silappatikaram (சிலப்பதிகாரம்) 1st of the Five Great Epics ~2nd century CE
Manimekalai (மணிமேகலை) 2nd of the Five Great Epics ~5th century CE
Seevaga Sinthamani (சீவகசிந்தாமணி) 3rd of the Five Great Epics ~10th century CE
Valayapathi (வளையாபதி) 4th of the Five Great Epics ~10th century CE
Kundalakesi (குண்டலகேசி) 5th of the Five Great Epics ~5th century CE
Tirukkural (திருக்குறள்) 1,330 couplets — ethics, governance, love ~300 BCE – 500 CE
Sangam Literature Akananuru, Purananuru, Natrinai, and more 300 BCE – 300 CE
Contemporary Tamil News, social media, government documents, spoken dialects Present

Capabilities

┌─────────────────────────────────────────────────────────┐
│              SiMaSeVaKuGa LLM Capabilities              │
├─────────────────────────────────────────────────────────┤
│  🧠  Tamil Language Understanding & Generation          │
│  💬  Conversational AI (chatbots, assistants)           │
│  🤖  Agentic AI — Tamil-native AI agents                │
│  🎙️  Voice-first, dialect-aware inference               │
│  📄  Document summarisation & translation               │
│  ⚖️  Legal & governance document processing             │
│  🏥  Healthcare advisory in Tamil                       │
│  🌾  Agricultural advisory (rural dialects)             │
│  🎓  Education & e-learning platforms                   │
│  🏛️  E-governance & citizen services                    │
└─────────────────────────────────────────────────────────┘

Architecture

SiMaSeVaKuGa LLM
├── Base Model        → Transformer-based, Tamil-first pre-training
├── vLLM Layer        → High-throughput inference (PagedAttention)
├── sLLM Variants     → Lightweight models for edge/mobile deployment
├── Tamil Tokenizer   → Custom BPE tokenizer for Tamil script
├── Voice Module      → Dialect-aware ASR + TTS
└── API Gateway       → REST + gRPC inference endpoints

Roadmap

Phase Timeline Milestone
Phase 1 — Foundation Months 1–12 Corpus assembly, tokenizer, base pre-training
Phase 2 — Acceleration Months 12–24 Fine-tuning, RLHF, public API beta
Phase 3 — Scale Months 24–36 Production deployment, edge models, voice API

Use Cases

Developers and enterprises can use SiMaSeVaKuGa LLM to build:

  • GenAI applications — Tamil content generation, summarisation, Q&A
  • Agentic AI systems — autonomous Tamil-language agents for workflows
  • Government services — citizen-facing Tamil AI for e-governance
  • Healthcare — Tamil diagnostic assistants and patient communication
  • Education — Tamil tutoring systems and adaptive learning
  • Agriculture — voice-based advisory for Tamil farmers

Getting Started

# Clone the repository
git clone https://github.com/tvkvolunteers/simasevakuga-llm.git
cd simasevakuga-llm

# Install dependencies
pip install -r requirements.txt

# Run inference (once model weights are available)
python inference.py --prompt "தமிழ்நாட்டின் தலைநகரம் எது?"

API (Coming Soon)

from simasevakuga import TamilLLM

llm = TamilLLM(model="simasevakuga-7b")

response = llm.generate(
    prompt="தமிழ்நாட்டின் வரலாற்றை சுருக்கமாக கூறு",
    max_tokens=512,
    dialect="standard"  # or "madurai", "coimbatore", "jaffna"
)

print(response.text)

Contributing

We welcome contributions from Tamil developers, linguists, researchers, and AI engineers worldwide.

1. Fork the repository
2. Create your feature branch: git checkout -b feature/your-feature
3. Commit your changes: git commit -m 'Add Tamil corpus segment'
4. Push to the branch: git push origin feature/your-feature
5. Open a Pull Request

Areas where we need help:

  • Tamil corpus curation and cleaning
  • Tokenizer improvements for Tamil script
  • Dialect data collection (Madurai, Coimbatore, Jaffna, Eelam Tamil)
  • Evaluation benchmarks for Tamil NLP
  • Fine-tuning datasets for specific domains

License

Apache License 2.0 — open for Tamil people and the world.

Acknowledgements

Built with pride by TVK Volunteers — an independent volunteer network supporting Thamizhaga Vettri Kazhagam and the people of Tamil Nadu.

உழைப்பு மக்களுக்கு • வெற்றி TVK-க்கு For the People. For TVK.

Links