QuixiAI/WizardLM-1.0-Uncensored-CodeLlama-34b

TEXT GENERATIONConcurrency Cost:2Model Size:34BQuant:FP8Ctx Length:32kPublished:Aug 29, 2023License:llama2Architecture:Transformer0.0K Open Weights Cold

QuixiAI/WizardLM-1.0-Uncensored-CodeLlama-34b is a 34 billion parameter language model built upon the CodeLlama-34b architecture, offering strong coding abilities. This model is a retraining of WizardLM-13B-V1.0, specifically fine-tuned with a filtered dataset to reduce refusals, avoidance, and bias, making it more compliant than its predecessor. It is designed for applications requiring a less censored response profile, while still acknowledging inherent ethical beliefs within the base LLaMA model. The model is optimized for general conversational tasks using Vicuna-1.1 style prompts, with a notable context length of 32768 tokens.

Loading preview...

Model Overview

QuixiAI/WizardLM-1.0-Uncensored-CodeLlama-34b is a 34 billion parameter language model based on the CodeLlama-34b architecture, providing it with robust coding capabilities. This model is a retrained version of WizardLM/WizardLM-13B-V1.0, utilizing a carefully filtered dataset to minimize refusals, avoidance, and inherent biases often found in large language models. While acknowledging that no model is "truly uncensored" due to the ethical beliefs embedded in its foundational architecture, this iteration aims to be significantly more compliant and less restrictive in its responses compared to its predecessor.

Key Characteristics

  • Base Model: CodeLlama-34b, inheriting strong programming and logical reasoning skills.
  • Uncensored Training: Fine-tuned with a filtered dataset to reduce refusal and bias, offering a more open response generation.
  • Prompt Format: Utilizes Vicuna-1.1 style prompts for interaction, expecting a USER: and ASSISTANT: structure.
  • Context Length: Supports a substantial context window of 32768 tokens, allowing for longer and more complex interactions.

Intended Use Cases

This model is suitable for developers and applications that require:

  • Reduced Censorship: For use cases where a less restrictive and more direct response is preferred, understanding the user's responsibility for generated content.
  • Coding Assistance: Leveraging its CodeLlama foundation for code generation, debugging, and technical explanations.
  • General Conversational AI: Engaging in broad conversational tasks with a focus on compliance and directness.

Users are reminded that an uncensored model carries no inherent guardrails, and they are responsible for any content generated and published.

Finetunes
1 models