Ishowbackup/gemma-4-E4B-it-uncensored

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 18, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Ishowbackup/gemma-4-E4B-it-uncensored is a 7.9 billion parameter instruction-tuned Gemma-4 model, derived from Google's original, with its refusal behavior significantly reduced. It utilizes a norm-preserving biprojected obliteration method to remove censorship while maintaining model quality. This model is optimized for use cases requiring direct answers without AI identity disclaimers or content refusals, making it suitable for applications needing uncensored responses.

Loading preview...

Ishowbackup/gemma-4-E4B-it-uncensored: Refusal-Free Gemma-4

This model is an uncensored variant of Google's gemma-4-E4B-it, specifically engineered to eliminate refusal behaviors. It achieves this through a novel "norm-preserving biprojected obliteration" method, which selectively removes refusal directions from the model's weights without degrading overall quality.

Key Capabilities & Differentiators

  • Significantly Reduced Refusals: Achieves a refusal rate of 0.7% across multiple independent datasets (686 prompts), down from 99% in the original model on specific tests. Most remaining "refusals" are false positives where the model still complies after a disclaimer.
  • Norm-Preserving Abliteration: Unlike standard projection methods, this technique guarantees that the model's weight magnitudes are preserved, ensuring no degradation in response quality (harmless response length ratio remains ~1.01).
  • Advanced Uncensoring Method: Employs per-layer refusal directions and a deterministic single-pass process, offering a more precise and efficient uncensoring approach compared to other methods like vanilla Heretic.
  • Gemma-4 Base: Built upon the google/gemma-4-E4B-it architecture, providing a strong foundation for general language understanding and generation.

Ideal Use Cases

  • Applications Requiring Direct Responses: Suitable for scenarios where the model needs to provide direct answers without ethical disclaimers or content refusals.
  • Research into Model Alignment & Safety: Can be used as a baseline or comparison for studying the effects of uncensoring techniques and their impact on model behavior.
  • Creative & Unrestricted Content Generation: Useful for tasks that benefit from a model less constrained by built-in refusal mechanisms.

Finetunes
2 models