usermma/LFM2.5-1.2B-MEGABRAIN-Thinking-Claude-Polaris-Deepseek-GLM-abliterated-failspy

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.2BQuant:BF16Context Size:32kPublished:Jun 5, 2026Architecture:Transformer0.0K Featherless Exclusive Cold

usermma/LFM2.5-1.2B-MEGABRAIN-Thinking-Claude-Polaris-Deepseek-GLM-abliterated-failspy is a 1.2 billion parameter language model derived from DavidAU/LFM2.5-1.2B-MEGABRAIN-Thinking-Claude-Polaris-Deepseek-GLM. It has been processed using the 'failspy' method via OBLITERATUS, an activation engineering tool designed to remove refusal behavior from models. This model is specifically engineered for applications requiring a language model with reduced refusal tendencies, offering a 32768 token context length.

Loading preview...

LFM2.5-1.2B-MEGABRAIN-Thinking-Claude-Polaris-Deepseek-GLM-abliterated-failspy

This model is a specialized 1.2 billion parameter language model, built upon the DavidAU/LFM2.5-1.2B-MEGABRAIN-Thinking-Claude-Polaris-Deepseek-GLM base. Its primary distinction lies in its processing through the failspy method, utilizing the OBLITERATUS tool. OBLITERATUS is an open-source activation engineering framework designed to modify language models by removing refusal behaviors.

Key Capabilities

  • Refusal Behavior Reduction: Engineered to exhibit reduced refusal tendencies compared to its base model, achieved via the failspy method.
  • Base Model Heritage: Inherits the core capabilities and architecture of the DavidAU/LFM2.5-1.2B-MEGABRAIN-Thinking-Claude-Polaris-Deepseek-GLM model.
  • Context Length: Supports a substantial context window of 32768 tokens.

Good For

  • Applications where mitigating model refusal is a critical requirement.
  • Research into activation engineering and model behavior modification.
  • Use cases demanding a 1.2B parameter model with a specific behavioral profile.

Finetunes
1 models