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

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.2BQuant:BF16Context Size:32kPublished:Feb 16, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

DavidAU/LFM2.5-1.2B-Thinking-SuperMinds-7x-Heretic-Uncensored-DISTILL is a 1.2 billion parameter LFM2.5 model fine-tuned for deep reasoning and uncensored output, featuring a 32768-token context length. This model was trained on seven specialized, high-reasoning datasets, with its reasoning capabilities completely replaced and optimized for compact yet detailed responses. It is a "Heretic" model, meaning it was de-censored prior to fine-tuning to ensure uninhibited content generation without refusals.

Loading preview...

Model Overview

DavidAU/LFM2.5-1.2B-Thinking-SuperMinds-7x-Heretic-Uncensored-DISTILL is a 1.2 billion parameter model based on the LFM2.5 architecture, specifically fine-tuned for enhanced reasoning and uncensored content generation. It leverages a 32768-token context window and was trained using Unsloth on local hardware at 16-bit precision.

Key Capabilities & Features

  • Deep Reasoning: The model's core reasoning capabilities have been entirely replaced and optimized through training on seven specialized, high-reasoning datasets. This results in compact, yet highly detailed and direct reasoning.
  • Uncensored Output ("Heretic" Model): This model was explicitly de-censored before fine-tuning, ensuring it generates content without refusals or built-in nanny-like restrictions. While it will generate sensitive content, users may need to provide specific directives (e.g., using slang terms) to achieve the desired graphic or explicit level.
  • Stable Reasoning: Reasoning performance is noted to be stable across a temperature range of 0.1 to 2.5.
  • Optimized for Specific Settings: Recommendations include using q5, q6, q8, or 16-bit precision, or Imatrix IQ3_M minimum. A repetition penalty of 1.05 to 1.1 is suggested, with lower temperatures (0.3-0.7) to prevent looping during thinking.

Recommended Use Cases

This model is particularly suited for applications requiring:

  • Unrestricted Content Generation: Ideal for creative writing, roleplay, or scenarios where content filtering is undesirable.
  • Detailed and Direct Reasoning: For tasks where concise yet thorough logical thought processes are beneficial.
  • Exploration of Sensitive Topics: When the ability to generate content on potentially controversial or explicit subjects without refusal is paramount.