TheBloke/Wizard-Vicuna-13B-Uncensored-SuperHOT-8K-fp16 is a 13 billion parameter language model, created by merging Eric Hartford's Wizard Vicuna 13B Uncensored with Kaio Ken's SuperHOT 8K. This model is provided in fp16 PyTorch format and features an extended context window of 8192 tokens, achieved through the SuperHOT 8K merge. It is designed for GPU inference and further conversions, offering an uncensored base for various applications.
Loading preview...
Model Overview
This model, TheBloke/Wizard-Vicuna-13B-Uncensored-SuperHOT-8K-fp16, is a 13 billion parameter language model derived from a merge of two distinct models: Eric Hartford's Wizard Vicuna 13B Uncensored and Kaio Ken's SuperHOT 8K. It is distributed in fp16 PyTorch format, suitable for GPU inference and as a base for further model conversions.
Key Capabilities & Features
- Extended Context Window: Integrates Kaio Ken's SuperHOT 8K, enabling an 8192-token context length during inference. This is activated by using
trust_remote_code=Truein Hugging Face Transformers. - Uncensored Base: Built upon Eric Hartford's Wizard Vicuna 13B Uncensored, which had alignment/moralizing responses removed from its training data. This provides a model without inherent guardrails, allowing for custom alignment.
- PyTorch fp16 Format: Optimized for GPU inference and serves as a flexible base for various quantization and conversion processes.
Use Cases
- Long Context Applications: Ideal for tasks requiring extensive contextual understanding, such as summarizing long documents, complex question-answering, or maintaining coherent dialogue over many turns.
- Custom Alignment & Fine-tuning: Provides a neutral, uncensored foundation for developers to implement their own specific alignment, safety, or ethical guidelines through further fine-tuning or RLHF (Reinforcement Learning from Human Feedback).
- Research & Development: Suitable for researchers exploring the effects of extended context and uncensored model behaviors.