OPEN-MYTHOS-2B

A Powerful Distilled Open-Source Model for Reasoning & Security Research

HOW TO RUN

Open-Mythos — Model Setup Guide

Quick Start

Step 1: Install Ollama

# macOS / Linux
curl -fsSL https://ollama.com/install.sh | sh

# Windows — download from https://ollama.com/download

Step 2: Pull the Model

ollama pull OPEN_AI_REASEARCH/Open-Mythos

Step 3: Run the Model

ollama run OPEN_AI_REASEARCH/Open-Mythos

You now have a working local LLM. Chat with it directly in the terminal.

---

PyTorch Weights (Advanced)

For fine-tuning, research, or loading in Python with full tensor access, you can convert the local Ollama GGUF model back into PyTorch format.

Convert GGUF to PyTorch

Download and run the official conversion tool:

# Linux / macOS
curl -sL https://raw.githubusercontent.com/bluerangers/OPEN-MYTHOS/main/gguf_to_pt.py | python3

Note: You must have torch installed and have pulled the model via Ollama first.

Use with Transformers

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-0.5B")
weights = torch.load("mythos_converted.pt")
model.load_state_dict(weights)

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-0.5B")
inputs = tokenizer("Hello, I am", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0]))

---

All-in-One Commands

macOS / Linux

# Install Ollama + pull model + convert to PyTorch — one line
curl -fsSL https://ollama.com/install.sh | sh && \
ollama pull OPEN_AI_REASEARCH/Open-Mythos && \
curl -sL https://raw.githubusercontent.com/bluerangers/OPEN-MYTHOS/main/gguf_to_pt.py | python3 && \
ollama run OPEN_AI_REASEARCH/Open-Mythos

---

Requirements

Component	Required	Notes
Ollama	Yes	For running the GGUF model
Python 3.8+	For .pt weights	Pre-installed on macOS/Linux
PyTorch	For .pt weights	pip install torch

Install PyTorch (if needed)

pip install torch

---

Model Details

Property	Value
Name	Open-Mythos
Base	Qwen2.5 architecture
Parameters	3 Billion
Formats	GGUF (Ollama)
Ollama	OPEN_AI_REASEARCH/Open-Mythos
License	Apache 2.0

---

Troubleshooting

“torch not found”

pip install torch

Ollama not starting

# Linux
sudo systemctl start ollama

# macOS — launch from Applications

Efficient. Intelligent. Fully Open.
Released April 2026

---

Overview

OPEN-MYTHOS-2B is a high-performance 2 billion parameter distilled model derived from our larger Mythos series.

Despite its compact size, this distilled model delivers exceptional reasoning, code understanding, and vulnerability detection capabilities — making frontier-level intelligence accessible on laptops, edge devices, and local servers.

MythOS — Our AI-native knowledge platform — is available here.

---

Key Features

• Distilled for Efficiency: Retains strong capabilities of much larger models while running efficiently on consumer hardware
• Excellent code analysis and vulnerability detection
• Strong multi-step reasoning and agentic abilities
• Fast inference (especially with quantization)
• Fully open weights and training methodology
• 128K context window support

---

Benchmarks

Vulnerability Detection & Bug Finding (Grok Distilled 2B)

Benchmark	Description	Mythos-2B Score	Comparison (Best 7B Model)	Notes
Zero-Day Simulation	Multi-stage attack planning	74.8%	71%	Strong performance for size
CVE Discovery Rate	Real-world CVE identification	82%	78%	Scanned OpenBSD & Linux modules
SWE-Bench Lite	Real GitHub issue resolution	48.6%	45%	Competitive with larger models
Memory Safety Bugs	Use-after-free, overflows, etc.	87.3%	82%	Excellent for a 2B model
Cryptographic Flaw Detection	Weak crypto & implementation issues	89%	84%	Very capable in audits

Efficiency Highlights

• Runs at ~45–60 tokens/sec on a single RTX 4090 (4-bit quantized)
• Fits comfortably in 4–6 GB VRAM (quantized)
• Can run on CPU + RAM (Apple Silicon M2/M3/M4 performs exceptionally well)
• Ideal for local security auditing, code review agents, and personal research

Full benchmark results → benchmarks/

---

Quick Start

”`bash

Hugging Face

pip install transformers