saidutta69/Qwen2.5-1.5B-Instruct-heretic

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 2, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

saidutta69/Qwen2.5-1.5B-Instruct-heretic is a 1.5 billion parameter instruction-tuned causal language model, a decensored variant of Qwen/Qwen2.5-1.5B-Instruct. Developed by RACER IS OP using Heretic v1.2.0, it suppresses refusal behavior via targeted weight edits to attention and MLP projections. This model is optimized for direct answers, making it suitable for local agents, roleplay, and research into alignment mechanics where refusal suppression is desired.

Loading preview...

Overview

This model, saidutta69/Qwen2.5-1.5B-Instruct-heretic, is a 1.5 billion parameter instruction-tuned language model based on the Qwen2.5 architecture. It was created by RACER IS OP using the Heretic v1.2.0 tool, which employs directional ablation (abliteration) to modify the base model's behavior. Unlike traditional fine-tuning, this method involves targeted weight edits to the attention output and MLP down-projections, effectively suppressing refusal behavior while largely preserving the original model's knowledge and instruction-following capabilities.

Key Differentiators

  • Decensored Output: Significantly reduces refusal rates (from 99/100 to 1/100 adversarial prompts) compared to the base model, ensuring more direct answers.
  • Targeted Modification: Achieves refusal suppression through precise weight edits, resulting in a low KL divergence (0.1607) from the original model's output distribution, indicating minimal impact on other capabilities.
  • Resource Efficient: At 1.5 billion parameters, it is designed to run efficiently on CPUs or low-VRAM GPUs, outperforming the 0.5B variant in reasoning and coherence.

Ideal Use Cases

This model is particularly well-suited for:

  • Local Agents: Deploying AI agents that require direct responses without refusal.
  • Roleplay: Creating more compliant and engaging roleplay scenarios.
  • Alignment Research: Studying refusal mechanics and the effects of alignment techniques.
  • Applications Blocked by Over-Refusal: Any use case where the base model's inherent refusal behavior is a hindrance.

Finetunes
1 models