🧠 IBM-Opus4.7-Obscure.Reasoner.3B.GGUF

Repository: WithinUsAI/IBM-Opus4.7-Obscure.Reasoner.3B.GGUF

---

🌌 Model Overview

IBM-Opus4.7-Obscure.Reasoner.3B.GGUF is a reasoning-specialized 3B language model created by WithIn Us AI and built from:

• IBM Granite 4.1 3B
• High-reasoning Opus-style distillation datasets
• Recursive analytical training methodologies
• Structured reasoning and decomposition tuning

The goal of this model is to push unusually deep reasoning behavior into a compact local model format while maintaining strong speed and accessibility for consumer hardware.

This model focuses heavily on:

• multi-step reasoning
• abstract analysis
• coding cognition
• recursive thought chains
• logical decomposition
• reflective response generation

Instead of behaving like a lightweight chat model, Obscure.Reasoner is tuned to operate more like a compact analytical engine designed for deep thinking tasks.

---

🧬 Base Model

Attribute	Value
Base Architecture	IBM Granite 4.1 3B
Format	GGUF
Parameter Class	~3B
Creator	WithIn Us AI
Primary Focus	High Reasoning
Inference Type	Local / Offline

---

⚡ Training Focus

This model was fine-tuned using:

• Opus-style reasoning distillation datasets
• high-complexity analytical prompts
• structured chain-of-thought style samples
• recursive reasoning patterns
• coding and logical decomposition tasks

Training emphasis prioritized:

• reasoning depth over shallow speed
• coherent multi-step answers
• analytical persistence
• reflective problem solving
• compact intelligence density

---

🧠 Behavioral Characteristics

Obscure.Reasoner tends to:

• think through problems step-by-step
• provide layered explanations
• analyze before concluding
• decompose abstract concepts
• perform well on recursive prompts
• sustain longer reasoning chains than typical small models

The model is especially effective for:

• coding assistance
• philosophical exploration
• AI cognition experiments
• prompt engineering
• local autonomous agents
• analytical writing
• logic-heavy tasks

---

🚀 Recommended Settings

Setting	Recommended
Temperature	0.65 – 0.85
Top-p	0.92 – 0.98
Top-k	30 – 60
Context Length	8192+
Repeat Penalty	1.05 – 1.12

For strongest reasoning:

• use structured prompts
• encourage step-by-step thinking
• ask decomposition-style questions
• avoid extremely short prompts

---

🖥️ Hardware Requirements

Quant	Approximate Memory
Q4_K_M	2–3 GB
Q5_K_M	3–4 GB
Q8_0	5–6 GB

Compatible with:

• llama.cpp
• LM Studio
• KoboldCpp
• Ollama
• Open WebUI
• local RAG systems
• lightweight AI agents

---

💻 Example llama.cpp Usage

llama-cli \
  -m IBM-Opus4.7-Obscure.Reasoner.3B.gguf \
  --ctx-size 8192 \
  --temp 0.72 \
  --top-p 0.95

---

🌠 Design Philosophy

Intelligence is not only scale. Intelligence is compression, structure, and reasoning density.

IBM-Opus4.7-Obscure.Reasoner.3B.GGUF was designed around the idea that a compact model can still exhibit surprisingly deep analytical behavior when trained on high-quality reasoning-focused datasets.

Rather than brute-force parameter count, the model emphasizes:

• cognitive efficiency
• structured reasoning
• analytical continuity
• compressed thought depth

A small lantern carrying a large flame. 🔦🧠

---

📚 Intended Use

Recommended for:

• local AI experimentation
• reasoning research
• coding assistance
• analytical prompting
• offline inference
• creative cognition systems
• recursive AI workflows

Not recommended for:

• factual certainty without verification
• legal advice
• medical advice
• safety-critical autonomous deployment

---

🙏 Acknowledgements

Special thanks to:

• IBM Granite researchers
• GGUF ecosystem developers
• llama.cpp contributors
• reasoning dataset creators
• open-source AI researchers
• the local inference community 🚀