A Proto-AGI (Prototype Artificial General Intelligence) system fine-tuned from TinyLlama-1.1B with specialized training in reasoning, tool calling, and Spanish language support.

Features

ReAct Cycle: THINK -> PLAN -> ACT -> OBSERVE -> REFLECT
Experiential Learning: Learns from past interactions
Metacognition: Self-evaluation for improvement
Web Search: Internet search capabilities
Multi-Agent: Collaboration between specialized agents (CODER, REVIEWER, RESEARCHER, PLANNER, TESTER)
Knowledge Base: RAG (Retrieval Augmented Generation)
Feedback Loop: Continuous improvement with user feedback
Tool Calling: Integrated tools for calculations, file operations, code execution
TTS Support: Text-to-Speech integration
Image Generation: Stable Diffusion integration
MCP Integration: Model Context Protocol support

Available Tools

Tool	Description
`calculate`	Mathematical calculations
`read_file`	Read files
`write_file`	Write files
`list_directory`	List directories
`execute_python`	Execute Python code
`web_search`	Search on internet
`fetch_url`	Get URL content
`research`	Deep research
`text_to_speech`	Convert text to speech
`generate_image`	Generate images

Operation Modes

CHAT: Casual conversation
TASK: Specific tasks with tools
RESEARCH: Deep information search
COLLABORATIVE: Multi-agent collaboration
LEARNING: Intensive learning mode

Usage

With Transformers

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("luepow/thau-agi-v2")
tokenizer = AutoTokenizer.from_pretrained("luepow/thau-agi-v2")

prompt = "<|system|>\nYou are THAU AGI v2, a helpful AI assistant.</s>\n<|user|>\nWhat is 25 * 4 + 100?</s>\n<|assistant|>\n"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

With Ollama

bash

ollama pull luepow/thau:agi-v2
ollama run luepow/thau:agi-v2

With Gradio Interface

bash

git clone https://github.com/luepow/thau.git
cd thau
pip install -r requirements.txt
python scripts/gradio_thau_ollama.py

Training Data

The model was fine-tuned on:

Programming tutorials (Python, JavaScript, Rust, Go, Java)
Mathematical reasoning
Tool calling patterns
Spanish language content
DevOps and cloud infrastructure
Agile methodologies
UX/CSS frameworks

Model Card

Base Model: TinyLlama/TinyLlama-1.1B-Chat-v1.0
Parameters: 1.1B
Context Length: 4096 tokens
Languages: English, Spanish
License: MIT

Credits

Developed with love for Thomas & Aurora.

THAU = THomas + AUrora

luepow/thau-agi-v2

THAU AGI v2 - Proto-AGI System