Introducing the Megha AI Model

A high-performance 0.6B parameter large language model designed specifically for agentic workflows, coding tasks, and multilingual support across 100+ languages.

Experience groundbreaking advancements in reasoning, instruction-following, and deep AI integration.

View on OllamaView on Hugging Face
Megha AI Model

Capabilities

Why choose the Megha Model?

Multilingual Support

100+ Languages

Built with strong capabilities for multilingual instruction following and translation, breaking down language barriers across the globe.

Thinking Mode

Advanced Thinking Mode

Toggle 'Thinking Mode' on or off dynamically. Megha can use its reasoning abilities to generate a chain of thought before providing complex answers.

Coding & Agents

Agentic & Coding Tasks

With a massive 32K context window, Megha is perfectly tuned to handle large codebases, deep reasoning, and autonomous agent workflows.

Technical Specifications

Under the hood of Megha

  • Model Type Causal Language Model
  • Total Parameters 0.6 Billion
  • Non-Embedding Parameters 0.44 Billion
  • Context Length 32,768 Tokens
  • Architecture 28 Layers, 16 Attention Heads (Q), 8 for KV

Integration & Deployment

How to use Megha in production

Run Locally with Ollama

You can instantly pull and run the Megha model on your local machine using the Ollama CLI. It's incredibly fast and requires minimal setup.

ollama run meghaos/megha

Python Integration (Hugging Face)

Integrate Megha directly into your Python apps using the standard Transformers library.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "MeghaOS/Megha"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

Production Deployment (vLLM & SGLang)

For high-throughput enterprise deployments, Megha is fully compatible with both vLLM and SGLang to instantly create an OpenAI-compatible API endpoint.

Using vLLM:
vllm serve MeghaOS/Megha --enable-reasoning --reasoning-parser deepseek_r1
Using SGLang:
python -m sglang.launch_server --model-path MeghaOS/Megha --reasoning-parser Megha