from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "luepow/thau-7b",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("luepow/thau-7b")

messages = [
    {"role": "system", "content": "You are THAU, a cognitive AI assistant."},
    {"role": "user", "content": "Explain Python decorators with examples."}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

With Ollama

ollama run luepow/thau-7b

Tool Calling

THAU supports JSON-based tool invocation:

<tool_call>{"name": "execute_python", "arguments": {"code": "print(2+2)"}}</tool_call>

Limitations

No vision/multimodal capabilities
No internal thinking tokens (uses prompting-based CoT)
Quality depends on prompt engineering for complex tasks

License

Apache 2.0

Citation

bibtex

@misc{thau-7b,
  author = {Luis Perez},
  title = {THAU 7B: Cognitive AI Assistant},
  year = {2024},
  publisher = {HuggingFace},
  url = {https://huggingface.co/luepow/thau-7b}
}

Acknowledgments

Qwen Team for the excellent base model
Anthropic's Claude for AI pair programming assistance
TinyLlama Team for inspiration