Llama 3.2 1B Uncensored Overview

This model, llama3.2_1b_2025_uncensored_v2, is a 1 billion parameter variant of Meta's Llama 3.2, fine-tuned by Carsen Klock. It distinguishes itself through its focus on providing uncensored responses and specialized capabilities in medical reasoning, mathematics problem-solving, and code generation. The model was trained using LoRA fine-tuning with Unsloth, across 79,263 steps.

Key Capabilities

• Uncensored Responses: Designed to engage with any topic and provide direct, honest answers without moral constraints. Users can activate this mode via a specific system prompt.
• Specialized Reasoning: Excels in complex domains such as medical reasoning and advanced mathematics, leveraging datasets like medical-o1-reasoning-SFT and math-gpt-4o-200k.
• Code Generation & Feedback: Proficient in generating code and providing feedback, trained on datasets like CodeFeedback-Filtered-Instruction and dolphin-coder.
• General Instruction Following: Maintains strong general instruction following capabilities from its diverse training data, including FineTome-100k.

Training Details

The model was fine-tuned on a combination of datasets including FineTome-100k (general instructions), orca-math-word-problems-200k and math-gpt-4o-200k (mathematics), CodeFeedback-Filtered-Instruction and dolphin-coder (code), Jenna-50K-Alpaca-Uncensored (uncensored conversations), and medical-o1-reasoning-SFT (medical reasoning). It supports a 32768 token context length and is available in GGUF (f16, q4_k_m) and merged 16-bit formats.

When to Use This Model

This model is particularly well-suited for applications requiring:

• Direct and unfiltered responses on a wide range of subjects.
• Assistance with complex mathematical problems.
• Code generation, debugging, and feedback.
• Medical information processing or reasoning tasks.
• General instruction following where a smaller, specialized model is preferred.