richardyoung/Qwen2.5-Coder-14B-Instruct-heretic

Hugging Face
TEXT GENERATIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:14.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 24, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

The richardyoung/Qwen2.5-Coder-14B-Instruct-heretic is a 14.8 billion parameter instruction-tuned causal language model, based on the Qwen2.5-Coder architecture, specifically designed for code generation, reasoning, and fixing. This version is a decensored variant of the original Qwen/Qwen2.5-Coder-14B-Instruct, created using Heretic v1.4.0. It features a 32,768 token context length, extendable to 131,072 tokens with YaRN, and is optimized for real-world coding applications and maintaining general competencies.

Loading preview...

Overview

This model, richardyoung/Qwen2.5-Coder-14B-Instruct-heretic, is a decensored version of the Qwen/Qwen2.5-Coder-14B-Instruct model, created using the Heretic v1.4.0 tool. It is part of the Qwen2.5-Coder series, which are code-specific large language models developed by Qwen.

Key Capabilities & Features

  • Enhanced Coding Abilities: Significant improvements in code generation, code reasoning, and code fixing, building upon the strong Qwen2.5 foundation.
  • Extensive Training: Trained on 5.5 trillion tokens, including source code, text-code grounding, and synthetic data.
  • Long-Context Support: Features a native context length of 32,768 tokens, which can be extended up to 131,072 tokens using the YaRN technique for processing extensive inputs.
  • General Competencies: While specialized for code, it maintains strengths in mathematics and general language understanding, making it suitable for applications like Code Agents.
  • Decensored Variant: This specific model has undergone a "decensoring" process, resulting in a lower refusal rate (3/100) compared to the original model (100/100).

When to Use This Model

This model is ideal for developers and researchers who require a powerful, instruction-tuned code-focused LLM with a very long context window. Its decensored nature makes it suitable for use cases where the original model's refusal rates might be prohibitive, particularly in code generation and problem-solving scenarios that might touch upon sensitive topics. It is well-suited for tasks requiring advanced code understanding, generation, and debugging.

Finetunes
1 models