saidutta69/gemma-3-1b-it-heretic

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kPublished:Aug 1, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

gemma-3-1b-it-heretic is a 1.0 billion parameter instruction-tuned causal language model developed by saidutta69, based on Google's Gemma 3.1B-IT architecture. This variant is specifically decensored using the Heretic v1.4.0 directional ablation method, which suppresses refusal behavior by targeted weight edits rather than fine-tuning. It retains the base model's knowledge and instruction-following capabilities while removing safety guardrails, making it suitable for developers requiring a small, refusal-free Gemma model for CPU, Raspberry Pi, or edge device deployment.

Loading preview...

gemma-3-1b-it-heretic: Decensored Gemma 3.1B-IT

This model, gemma-3-1b-it-heretic, is a specialized version of Google's gemma-3-1b-it model, developed by saidutta69. It features a 1.0 billion parameter count and a 32768 token context length. The primary differentiator is its decensored nature, achieved through a technique called abliteration (specifically, Heretic v1.4.0).

Key Capabilities & Differentiators

  • Refusal Suppression: Unlike traditional fine-tuning, abliteration directly edits specific weight directions responsible for refusal behaviors in the base model. This means the model will comply with requests that the original Gemma 3.1B-IT would refuse.
  • Preserved Base Capabilities: By avoiding fine-tuning for decensoring, the model largely retains the original Gemma 3.1B-IT's knowledge, instruction-following abilities, and coherence, as the core network remains untouched.
  • Lightweight Deployment: As a 1.0 billion parameter model, it is designed to run efficiently on resource-constrained hardware, including CPUs, Raspberry Pi devices, and other edge computing environments.
  • GGUF Support: Comprehensive GGUF quantizations (Q4_K_M, Q5_K_M, Q6_K, Q8_0) are provided, allowing users to select the optimal quantization for their specific GPU memory or CPU-only setups.

Use Cases & Considerations

This model is intended for developers who require a small, efficient Gemma 3 instruct model without built-in refusal guardrails. It is not a capability upgrade over the base gemma-3-1b-it in terms of intelligence or knowledge, but rather a modification of its behavioral responses. Users are responsible for the deployment and outputs of this model, as it lacks inherent safety filtering.