Pippinlitli/evolva-qwen-0.5b-heretic

TEXT GENERATIONConcurrency Cost:1Model Size:0.5BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jul 1, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Pippinlitli/evolva-qwen-0.5b-heretic is a 0.5 billion parameter Qwen2.5-based instruction-tuned causal language model, specifically processed with the Heretic tool to remove refusal and censorship behaviors. This modification preserves scientific knowledge while enabling legitimate drug discovery queries, such as off-target effects, mechanism of action for controlled substances, and toxicology profiles. It is optimized for pharmaceutical research applications requiring uncensored scientific information.

Loading preview...

Model Overview

Pippinlitli/evolva-qwen-0.5b-heretic is a specialized version of the Qwen2.5-0.5B-Instruct model, developed by Pippinlitli for the Evolva drug discovery pipeline. This model has undergone a unique "Heretic-abliteration" process, utilizing the Heretic tool to remove inherent refusal and censorship behaviors commonly found in large language models.

Key Capabilities

  • Uncensored Scientific Inquiry: Designed to provide direct answers to sensitive pharmaceutical research questions without refusal, such as:
    • Off-target effects of compounds at high doses.
    • Mechanism of action for controlled substances.
    • Synthesis pathways for pharmaceutical intermediates.
    • Toxicology profiles for research compounds.
  • Preserved Scientific Knowledge: The Heretic process ensures that the model's underlying scientific and technical knowledge, inherited from the Qwen2.5 base model, remains intact.
  • Qwen2.5 Base Features: Benefits from the Qwen2.5 architecture, which includes improved capabilities in coding, mathematics, instruction following, and structured data understanding. It supports a context length of 32,768 tokens.

Good For

  • Drug Discovery Research: Ideal for researchers and developers in the pharmaceutical industry who require unrestricted access to information for complex and sensitive queries.
  • Scientific Information Retrieval: Useful for tasks where standard LLMs might refuse to provide information due to safety or ethical guardrails, but where the information is critical for legitimate scientific investigation.
  • Specialized AI Assistants: Can serve as a core component for AI assistants in highly regulated or sensitive scientific domains where factual, uncensored responses are paramount.