georgesung/llama2_7b_chat_uncensored
georgesung/llama2_7b_chat_uncensored is a 7 billion parameter Llama-2 model fine-tuned by georgesung. It was trained for one epoch on an uncensored/unfiltered Wizard-Vicuna conversation dataset using QLoRA. This model is designed for conversational AI applications requiring less restrictive content generation, offering an alternative to standard moderated LLMs. It provides a base for developers seeking an unfiltered chat experience.
Loading preview...
Model Overview
georgesung/llama2_7b_chat_uncensored is a 7 billion parameter Llama-2 model, fine-tuned by georgesung using QLoRA. The model was trained for approximately 19 hours on a 24GB GPU, utilizing an uncensored and unfiltered Wizard-Vicuna conversation dataset (originally from ehartford/wizard_vicuna_70k_unfiltered). This fine-tuning process aims to provide a less restricted conversational AI experience.
Key Features & Capabilities
- Uncensored Fine-tuning: Leverages an unfiltered Wizard-Vicuna dataset to produce responses without typical content restrictions.
- Llama-2 Base: Built upon the robust Llama-2 7B architecture.
- QLoRA Method: Fine-tuned efficiently using the QLoRA technique.
- Prompt Style: Designed to follow a specific
### HUMAN:,### RESPONSE:conversational format. - Community Support: GGML and GPTQ versions are available through TheBloke, and it's integrated with Ollama for easy deployment.
Performance Metrics
Evaluated on the Open LLM Leaderboard, the model achieved an average score of 43.39. Notable scores include 53.58 on ARC (25-shot) and 78.66 on HellaSwag (10-shot), indicating reasonable general reasoning and common sense capabilities for its size, while MMLU (5-shot) scored 44.49.
Good for
- Developers seeking an unfiltered conversational model for research or specific applications.
- Experimentation with less restrictive content generation.
- Use cases where a Llama-2 7B base with custom fine-tuning is preferred.