richardyoung/fable-qwen2.5-3b-agentic-merged-heretic

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 26, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The richardyoung/fable-qwen2.5-3b-agentic-merged-heretic is a 3.1 billion parameter Qwen2.5-3B-Instruct model, developed by devxyasir and further modified by Richard Young. This model is a decensored version of its base, optimized for reduced refusals in responses. It is fine-tuned for general instruction-following tasks, offering a 32K context length.

Loading preview...

Model Overview

This model, richardyoung/fable-qwen2.5-3b-agentic-merged-heretic, is a 3.1 billion parameter instruction-tuned language model based on the Qwen2.5 architecture. It is a decensored variant of devxyasir/fable-qwen2.5-3b-agentic-merged, created using the Heretic v1.4.0 project. The original model was developed by devxyasir and fine-tuned from unsloth/qwen2.5-3b-instruct-unsloth-bnb-4bit using Unsloth for faster training.

Key Differentiators

  • Decensored Output: A primary feature of this model is its significantly reduced refusal rate compared to its base model. Performance metrics indicate a refusal rate of 3/100, in contrast to 96/100 for the original devxyasir/fable-qwen2.5-3b-agentic-merged.
  • Reproducibility: The model's creation process is reproducible, with detailed instructions available in the reproduce/README.md directory.
  • Abliteration Parameters: Specific abliteration parameters were applied during its modification, including adjustments to direction_index, attn.o_proj weights, and mlp.down_proj weights.

Use Cases

This model is suitable for applications requiring a less restrictive instruction-following language model, particularly where the base model's high refusal rate might be a limitation. Its 32K context length supports handling longer prompts and generating more extensive responses.

Finetunes
1 models