BlossomsAI/Qwen2.5-Coder-14B-Instruct-Uncensored

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:14.8BQuant:FP8Ctx Length:32kLicense:mitArchitecture:Transformer0.0K Open Weights Warm

BlossomsAI/Qwen2.5-Coder-14B-Instruct-Uncensored is an instruction-tuned language model developed by BlossomsAI, based on the Qwen2.5-Coder-14B-Instruct architecture. This model is specifically optimized to remove refusal behaviors through directional intervention techniques. It is designed for resource-efficient deployment, with pre-quantized versions available, making it suitable for code generation tasks where uninhibited responses are desired.

Loading preview...

Overview

BlossomsAI/Qwen2.5-Coder-14B-Instruct-Uncensored is a specialized variant of the Qwen2.5-Coder-14B-Instruct model, developed by BlossomsAI. Its primary distinction lies in the removal of refusal behaviors, achieved through the application of directional intervention techniques. This modification aims to provide a more unconstrained response generation, particularly beneficial for coding-related instructions.

Key Capabilities

  • Behavior Modification: Implements directional intervention to eliminate refusal behaviors, allowing for more direct and uninhibited responses.
  • Resource Efficiency: The model is fully quantized, with pre-quantized versions available in GGUF format, facilitating resource-efficient deployment on various hardware.
  • Foundation: Built upon the robust Qwen2.5-Coder-14B-Instruct architecture, inheriting its code generation capabilities.

Good For

  • Code Generation: Ideal for scenarios requiring a coding assistant that provides direct answers without refusal, even to potentially sensitive or complex prompts.
  • Research into Model Behavior: Useful for researchers studying model alignment and the effects of directional intervention on LLM responses.
  • Edge Deployment: The quantized versions make it suitable for deployment on devices with limited computational resources.