llama2_7b_chat_uncensored
| Property | Value |
|---|---|
| Parameter Count | 6.74B |
| License | Other |
| Training Time | 19 hours |
| Hardware Used | 24GB NVIDIA A10G GPU |
| Tensor Type | F32 |
What is llama2_7b_chat_uncensored?
llama2_7b_chat_uncensored is a fine-tuned version of the Llama-2 7B model, specifically designed to provide less restricted responses. The model was trained using QLoRA technology on an uncensored Wizard-Vicuna conversation dataset, making it more suitable for applications requiring unrestricted dialogue capabilities.
Implementation Details
The model utilizes the QLoRA fine-tuning approach, trained for one epoch on a 24GB NVIDIA A10G GPU. It implements a specific prompt structure using "### HUMAN:" and "### RESPONSE:" format for interactions. The model is available in multiple versions including fp16 HuggingFace, GGML, and GPTQ formats.
- Trained on georgesung/wizard_vicuna_70k_unfiltered dataset
- Compatible with Ollama platform
- Includes comprehensive evaluation metrics
- Supports multiple deployment options
Core Capabilities
- ARC (25-shot): 53.58% accuracy
- HellaSwag (10-shot): 78.66% accuracy
- MMLU (5-shot): 44.49% accuracy
- TruthfulQA (0-shot): 41.34% accuracy
- Winogrande (5-shot): 74.11% accuracy
Frequently Asked Questions
Q: What makes this model unique?
This model stands out due to its uncensored training approach and the use of QLoRA fine-tuning, making it more suitable for applications requiring unrestricted responses while maintaining the core capabilities of Llama-2.
Q: What are the recommended use cases?
The model is best suited for chat applications requiring more natural and unrestricted conversations, research purposes, and applications where standard content filters might be too restrictive. However, users should implement their own safety measures as needed.
