WithinUsAI/IBM4.1-Unnoticed.Thinker.Uncensored-3B.gguf

🧠 Model Card

IBM4.1-Unnoticed.Thinker.Uncensored-3B (GGUF)

Repository: WithinUsAI Format: GGUF Base Model: IBM Granite 4.1 3B (inferred) Architecture Type: Transformer-based LLM Parameter Size: ~3 Billion

✨ Overview

IBM4.1-Unnoticed.Thinker.Uncensored-3B is a lightweight, reasoning-oriented language model distributed in GGUF format for efficient local inference.

This model is designed with a focus on:

• —🧠 Structured thinking / reasoning
• —🔓 Uncensored response behavior
• —⚡ Local deployment efficiency (llama.cpp / GGUF stack)

It belongs to a growing class of “uncensored” models, which aim to reduce refusal rates and increase response completeness compared to standard aligned models. ([Hugging Face][1])

🧬 Model Characteristics

🧪 Training & Origin

• —Base model derived from IBM Granite 4.1 3B
• —Modified and/or fine-tuned by WithinUsAI
• —Converted to GGUF format for compatibility with:

• —llama.cpp
• —LM Studio
• —Ollama (via conversion)

⚙️ Usage

🖥️ Run with llama.cpp

./main -m IBM4.1-Unnoticed.Thinker.Uncensored-3B.gguf -p "Explain recursion simply"

🧪 Recommended Settings

• —Temperature: 0.6 – 0.8
• —Top-p: 0.85 – 0.95
• —Top-k: 20 – 50

These settings help balance:

• —🧠 coherent reasoning
• —🎲 creative exploration

🧭 Behavior Notes

This model is uncensored, meaning:

• —It may respond to prompts that other models decline
• —It prioritizes completeness over restriction
• —It may produce:

• —raw or unfiltered outputs
• —speculative or unsafe content

highlight that “uncensored” variants have **zero refusals and full response

license: apache-2.0 tags:

• —language
• —granite-4.1 --- Capabilities
• —Summarization
• —Text classification
• —Text extraction
• —Question-answering
• —Retrieval Augmented Generation (RAG)
• —Code related tasks
• —Function-calling tasks
• —Multilingual dialog use cases
• —Fill-In-the-Middle (FIM) code completions