Banner by Croissant

N1 - A Chain-of-Thought Language Model

N1 is a small, experimental Chain-of-Thought (COT) model based on the LLaMA architecture, developed by GoofyLM.

Model Details

• Architecture: LLaMA-based
• Parameter Count: 135M
• Training Data: Closed-source dataset
• Special Features: Chain-of-Thought reasoning capabilities
• Note: The model often shows "schizophrenia"
• Note: You may need to add this Jinja to the model:

{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system
You are a helpful AI assistant named N1, trained by GoofyLM<|im_end|>
' }}{% endif %}{{'<|im_start|>' + message['role'] + '
' + message['content'] + '<|im_end|>' + '
'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
' }}{% endif %}

• ONNX available at onnx-community/N1-ONNX.

Intended Use

This model is designed for text generation tasks with a focus on reasoning through problems step-by-step (using its Chain-of-Thought).

Limitations

• Small parameter size may limit reasoning capabilities
• May produce unstable or inconsistent outputs
• Not suitable for production use without further testing

---

Usage

The model can be loaded using the following:

Transformers:

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("GoofyLM/N1")
tokenizer = AutoTokenizer.from_pretrained("GoofyLM/N1")

llama-cpp-python:

from llama_cpp import Llama

llm = Llama.from_pretrained(
    repo_id="GoofyLM/N1-quant",
    filename="N1_Q8_0.gguf",
)

Ollama:

ollama run hf.co/GoofyLM/N1-quant:Q8_0