sonic-coder/Qwen2-0.5B-Instruct-Abliterated

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

sonic-coder/Qwen2-0.5B-Instruct-Abliterated is a 0.5 billion parameter instruction-tuned causal language model based on the Qwen2 architecture. This model has been 'abliterated' using a specific procedure to modify its behavior, incorporating additional data related to harmful behaviors. It is designed for use cases where a modified Qwen2-0.5B-Instruct model with altered response characteristics is desired, particularly in safety research or content moderation contexts.

Loading preview...

Model Overview

sonic-coder/Qwen2-0.5B-Instruct-Abliterated is a 0.5 billion parameter instruction-tuned model derived from the Qwen2-0.5B-Instruct architecture. This version has undergone an 'abliteration' process, which involves modifying its training or fine-tuning data to influence its response generation, particularly concerning harmful content.

Key Characteristics

  • Base Model: Built upon the Qwen2-0.5B-Instruct foundation, offering a compact yet capable language model.
  • Abliteration Process: The model was modified using a procedure similar to that employed by augmxnt/Qwen2-7B-Instruct-deccp. This process involved adding specific lines from the mlabonne/harmful_behaviors dataset to a harmful.txt file, aiming to alter the model's handling of sensitive or harmful prompts.
  • Context Length: Supports a substantial context length of 32768 tokens, allowing for processing longer inputs.

Potential Use Cases

  • Safety Research: Investigating the effects of specific data modifications on model behavior, especially concerning harmful content generation.
  • Content Moderation Prototyping: Experimenting with models that have been intentionally exposed to or modified against harmful behaviors.
  • Comparative Analysis: Studying how 'abliteration' impacts the performance and safety alignment of instruction-tuned models compared to their original versions.