Justbackup/Llama3.3-8B-Instruct-Thinking-Heretic-Uncensored-Claude-4.5-Opus-High-Reasoning
Justbackup/Llama3.3-8B-Instruct-Thinking-Heretic-Uncensored-Claude-4.5-Opus-High-Reasoning is an 8 billion parameter Llama 3.3-based instruction-tuned model with a 128k context window. Developed by Justbackup, it has been de-censored and further trained on a Claude 4.5-Opus High Reasoning dataset using Unsloth. This model functions as an instruct/thinking hybrid, designed to automatically activate a 'thinking' process for complex prompts while remaining fully uncensored for diverse content generation.
Loading preview...
Model Overview
Justbackup/Llama3.3-8B-Instruct-Thinking-Heretic-Uncensored-Claude-4.5-Opus-High-Reasoning is an 8 billion parameter model based on a never-publicly-released Llama 3.3 source, adjusted for a 128k context window. This model has undergone a unique two-stage modification process: first, it was "Heretic'ed" (de-censored to significantly reduce refusals), and then fine-tuned for three epochs using Unsloth with a high-quality Claude 4.5-Opus High Reasoning dataset. The result is an instruct/thinking hybrid model that is fully uncensored and capable of advanced reasoning.
Key Capabilities and Features
- Hybrid Functionality: Operates as both an instruction-following model and a 'thinking' model, automatically activating an internal thought process for complex queries.
- Uncensored Output: Significantly reduced content refusals (14/100) compared to original models, allowing for generation of sensitive or explicit content when directed.
- High Reasoning: Enhanced reasoning capabilities derived from training on the Claude 4.5-Opus High Reasoning dataset.
- Extended Context: Supports a substantial context length of 128k tokens, suitable for long-form content generation and complex tasks.
- Adaptive Prompting: Recognizes specific phrases/words to automatically trigger its 'thinking' mode, while direct instructions can bypass this for more straightforward responses.
Usage Notes
While uncensored, the model may require explicit directives (e.g., including specific slang or graphic terms) to generate content at expected explicit or graphic levels. Suggested settings include a temperature of 0.7, repetition penalty of 1.05, top_p of 0.95, min_p of 0.05, and top_k of 40. It is recommended to use a minimum context window of 4k, ideally 8k+, and to avoid system prompts as thinking tags are self-generated.