Justbackup/LFM2.5-1.2B-Thinking-SuperMinds-7x-Heretic-Uncensored-DISTILL

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.2BQuant:BF16Context Size:32kPublished:Aug 20, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The Justbackup/LFM2.5-1.2B-Thinking-SuperMinds-7x-Heretic-Uncensored-DISTILL model is a 1.2 billion parameter LFM2.5 fine-tune, developed by Justbackup, specifically engineered for deep reasoning and uncensored output. It features a 32768 token context length and has been trained on seven specialized, high-reasoning datasets. This model excels at generating detailed, compact reasoning and provides uncensored responses, making it suitable for applications requiring direct and unfiltered content generation.

Loading preview...

Model Overview

Justbackup/LFM2.5-1.2B-Thinking-SuperMinds-7x-Heretic-Uncensored-DISTILL is a 1.2 billion parameter LFM2.5 model, fine-tuned by Justbackup using Unsloth at 16-bit precision. Its core differentiator is a completely replaced and enhanced reasoning capability, achieved through distillation with specialized datasets. The model was trained on seven high-reasoning datasets to produce compact yet highly detailed reasoning.

Key Capabilities

  • Enhanced Reasoning: The model's reasoning capabilities have been entirely re-engineered, affecting general operation, output generation, and benchmarks.
  • Uncensored Output: This is a "Heretic" model, meaning it is fully uncensored. It was decensored before fine-tuning to ensure consistent, unfiltered responses without refusals.
  • Extended Context: Features a 32768 token context window.
  • Temperature Stability: Reasoning remains stable across a wide temperature range of 0.1 to 2.5.

Optimal Usage

  • Recommended Quantization: Strongly suggests q5, q6, q8, 16-bit precision, or Imatrix IQ3_M minimum for best performance.
  • Repetition Penalty: Use a repetition penalty between 1.05 and 1.1.
  • Looping Mitigation: If looping occurs during thinking, lower the temperature to 0.3-0.7.
  • Smoothing Factor: For smoother operation in interfaces like KoboldCpp, oobabooga, or Silly Tavern, set the "Smoothing_factor" to 1.5.

Good For

  • Applications requiring deep, detailed, and compact reasoning.
  • Use cases where uncensored and direct content generation is critical.
  • Scenarios demanding a model that will not refuse requests, even for sensitive topics, provided appropriate directives are given.

For advanced settings and optimal performance guides, refer to the Maximizing Model Performance page.

Finetunes
1 models