canyrtcn/Gemma_E4B_Kizagan_Abliterated

VISIONConcurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 7, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

canyrtcn/Gemma_E4B_Kizagan_Abliterated is a 7.5 billion parameter (4B effective PLE) Gemma 4 E4B-it based language model, developed by canyrtcn, optimized for Turkish reasoning tasks. This model has undergone an 'abliteration' process to remove refusal behaviors, allowing it to generate unfiltered content. It supports a 131,072 token context length and is primarily intended for applications requiring direct, unrefused responses in Turkish.

Loading preview...

Model Overview

canyrtcn/Gemma_E4B_Kizagan_Abliterated is a Gemma 4 E4B-it based language model, fine-tuned for Turkish reasoning. It builds upon AlicanKiraz0/Kizagan-E4B-Turkish-Reasoning-Model, with a key modification: the application of an "abliteration" process. This process specifically targets and removes the model's refusal behaviors, enabling it to generate unfiltered content based on prompts.

Key Capabilities

  • Turkish Language Optimization: Primarily designed and optimized for Turkish, with some English language support.
  • Refusal Behavior Removal: Abliterated to eliminate content filtering tendencies, providing direct responses.
  • Gemma 4 E4B Base: Utilizes the Gemma 4 E4B-it architecture, featuring 7.5 billion total parameters (4 billion effective PLE).
  • Extended Context Window: Supports a substantial context length of 131,072 tokens.
  • Apache 2.0 License: Released under Apache License 2.0, subject to Gemma 4 terms.

Use Cases and Considerations

This model is suitable for applications where unfiltered, direct responses in Turkish are desired. Developers should be aware that due to the removal of refusal behaviors, the model may produce content that would typically be filtered by other models. Therefore, users bear the responsibility for the generated output. It is important to note its focus on Turkish; performance in other languages is not guaranteed. As a smaller model, it may not match the performance of larger models in highly specialized or extremely long-context scenarios, and like all LLMs, it is susceptible to hallucinations.

Finetunes
2 models