Ruliad logoRRuulliiaadd
Contact
← All models
Open weights· Nov 2024

DeepThought-8B

A small, capable reasoning model built on LLaMA-3.1 8B. Transparent step-by-step reasoning with test-time compute scaling.

Weights on Hugging Face
Specifications
Parameters
8B
License
Apache 2.0
Base
LLaMA-3.1 8B
VRAM
16GB+
Model IDruliad/deepthought-8b-llama-v0.01-alpha
Overview

A small model that shows its reasoning.

DeepThought-8B is a reasoning model built on LLaMA-3.1 8B, designed to make AI reasoning more transparent and controllable. It is small enough to run on a single consumer GPU.

The model breaks down its thinking into clear, distinct, documented steps. It outputs its reasoning process in a structured JSON format, making it easier to understand and validate its decision-making process.

Each reasoning chain follows a structured progression: problem understanding, data gathering, analysis, calculation (when applicable), verification, conclusion drawing, and implementation.

Using it

Key features

  • Transparent reasoning: Step-by-step documentation of the thought process in structured JSON
  • Programmable approach: Customizable reasoning patterns without model retraining
  • Test-time compute scaling: Flexible reasoning depth based on task complexity
  • Efficient scale: Runs on consumer GPUs with 16GB+ VRAM
  • Structured output: JSON-formatted reasoning chains for easy integration

Example output

The model provides structured reasoning in JSON format. Each reasoning chain follows a progression through problem understanding, data gathering, analysis, calculation, verification, conclusion, and implementation.

{
  "step": 1,
  "type": "problem_understanding",
  "thought": "Understanding the user's objective for the task."
}

Quick start

Install the required dependencies:

pip install torch transformers
# Optional: Flash Attention 2 for better performance
pip install flash-attn

Load and run the model:

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "ruliad/deepthought-8b-llama-v0.01-alpha"

tokenizer = AutoTokenizer.from_pretrained(
    model_name,
    add_bos_token=False,
    trust_remote_code=True,
    padding="left",
    torch_dtype=torch.bfloat16,
)

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    attn_implementation="flash_attention_2",
    use_cache=True,
    trust_remote_code=True,
)

Requirements

  • Python 3.6+
  • PyTorch
  • Transformers library
  • 16GB+ VRAM
  • Optional: Flash Attention 2 for improved performance

Known limitations

  • Complex mathematical reasoning
  • Long-context processing
  • Edge case handling

Citation

@misc{Deepthought2024,
  author = {Ruliad},
  title = {Deepthought-8B: A Small and Capable Reasoning Model},
  year = {2024},
  publisher = {Ruliad}
}