georgesung/open_llama_7b_qlora_uncensored

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:4kPublished:Jul 2, 2023License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

georgesung/open_llama_7b_qlora_uncensored is a fine-tuned OpenLLaMA-7B model, developed by georgesung, optimized for uncensored and unfiltered conversational AI. This 7 billion parameter model was fine-tuned using QLoRA on a Wizard-Vicuna conversation dataset. It is designed for applications requiring open-ended, unrestricted dialogue generation.

Loading preview...

Model Overview

This model, georgesung/open_llama_7b_qlora_uncensored, is a fine-tuned version of the OpenLLaMA-7B base model. Developed by georgesung, its primary differentiator is its training on an uncensored/unfiltered Wizard-Vicuna conversation dataset, specifically ehartford/wizard_vicuna_70k_unfiltered.

Key Characteristics

  • Base Model: OpenLLaMA-7B.
  • Fine-tuning Method: QLoRA (Quantized Low-Rank Adaptation).
  • Training Data: Uncensored/unfiltered Wizard-Vicuna conversation dataset.
  • Training Duration: Approximately 18 hours on a 24GB NVIDIA A10G GPU for one epoch.
  • Prompt Style: Utilizes a specific ### HUMAN: and ### RESPONSE: format for conversational turns.

Use Cases

This model is particularly suited for applications requiring:

  • Unrestricted Conversational AI: Generating responses without inherent content filters.
  • Exploration of Dialogue Systems: Research and development in open-domain chat.

Additional Resources

  • The training code used for this model is available on GitHub.
  • A Gradio chat application demo is provided for interactive use.
  • A blog post detailing the fine-tuning process is available here.