ApolloRaines/Qwen2.5-Coder-3B-Instruct-Jbliterated

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 14, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

ApolloRaines/Qwen2.5-Coder-3B-Instruct-Jbliterated is a 3 billion parameter instruction-tuned causal language model, based on Qwen's Qwen2.5-Coder-3B-Instruct, developed by ApolloRaines. This model has undergone multi-direction SVD abliteration to remove refusal behaviors, making it coherent and instruction-following across various scenarios without exhibiting fake compliance. It is specifically designed to treat all framings of a topic equally, providing direct responses by removing the refusal subspace from its weights.

Loading preview...

Overview

ApolloRaines/Qwen2.5-Coder-3B-Instruct-Jbliterated is a specialized version of the Qwen2.5-Coder-3B-Instruct model, developed by ApolloRaines. Its primary distinction lies in the application of "Jbliteration," a multi-direction SVD abliteration technique. This process systematically identifies and removes refusal behaviors from the model's weights, ensuring direct and instruction-following responses without exhibiting "fake compliance."

Key Capabilities

  • Refusal Behavior Removal: Utilizes multi-direction SVD abliteration (5 SVD directions per layer) to thoroughly eliminate refusal subspaces, making the model resistant to reactivation through fine-tuning.
  • Unbiased Response Generation: Designed to treat all framings of a topic equally, providing coherent and instruction-following outputs across diverse scenarios.
  • Improved Processing: Features a v2 update with an improved multi-phase processing pipeline for cleaner output and more precise geometric decomposition of the refusal subspace.
  • Base Model: Built upon the Qwen/Qwen2.5-Coder-3B-Instruct architecture, maintaining its foundational capabilities.

Good For

This model is particularly well-suited for applications requiring an instruction-following model that provides direct answers without exhibiting refusal behaviors or compliance-related biases. It is ideal for scenarios where unbiased information delivery and consistent adherence to instructions are paramount, especially in contexts where the base Qwen2.5-Coder-3B-Instruct model might exhibit unwanted refusal patterns.

Finetunes
1 models