MANAS-Base

MANAS is a small, efficient conversational AI model. It answers questions, does arithmetic, and chats — natively, without external tools.

Author: @k_k_kaundal on X

Files

• manas-base.pt — the model weights (PyTorch checkpoint).
• tokenizer.model / tokenizer.vocab — the SentencePiece tokenizer.

Capabilities

• 💬 Natural conversation
• ➕ Arithmetic (addition, subtraction, multiplication, division)
• 🌍 World facts (capitals, science, history, …)
• ❓ Question answering

Quick start

import torch
from huggingface_hub import hf_hub_download

# download weights + tokenizer
model_path = hf_hub_download("kaundalvip/manas-base", "manas-base.pt")
tok_path   = hf_hub_download("kaundalvip/manas-base", "tokenizer.model")

# load (requires the manas package)
ckpt = torch.load(model_path, map_location="cpu")
# ... build the model from ckpt["cfg"] and load ckpt["model"] ...

See the model in action:

You: Hi!                      → MANAS: Hello! How can I help you today?
You: What is 11 * 11?         → MANAS: 11 * 11 = 121
You: What is the capital of Italy? → MANAS: The capital of Italy is Rome.
You: What is photosynthesis?  → MANAS: Photosynthesis is how plants convert sunlight, water, and carbon dioxide into energy and oxygen.

Notes

• Lightweight and CPU-friendly.
• Best at short, direct questions and everyday conversation.

Connect

Built by K K Kaundal — follow on X: @k_k_kaundal