TheBloke/wizardLM-7B-HF

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Apr 26, 2023License:otherArchitecture:Transformer0.1K Cold

TheBloke/wizardLM-7B-HF is a 7 billion parameter instruction-following large language model, created by merging WizardLM delta weights with the original Llama7B model. This model is specifically designed to improve LLM performance through Evol-Instruct, a method that uses LLMs to automatically generate diverse instructions. It excels at handling high-complexity instructions, making it suitable for advanced instruction-following tasks.

Loading preview...

WizardLM-7B-HF Overview

WizardLM-7B-HF is an instruction-following large language model, derived from the Llama7B architecture. This 7 billion parameter model was created by applying WizardLM's delta weights to the base Llama7B model, with the merged model then converted to float16 for a standard 7B model size. The core innovation behind WizardLM is Evol-Instruct, a novel method that leverages LLMs themselves to automatically generate a wide range of open-domain instructions with varying difficulty levels and skill requirements.

Key Capabilities

  • Advanced Instruction Following: Designed to enhance LLM performance by training on a diverse, automatically generated instruction set.
  • High-Complexity Task Handling: Demonstrates strong performance on instructions with high complexity, outperforming some models like ChatGPT in specific complex scenarios.
  • Llama-based Architecture: Benefits from the foundational capabilities of the Llama7B model.

Good For

  • Developers seeking an instruction-following model optimized for complex prompts.
  • Research into instruction generation and LLM self-improvement techniques.
  • As a base model for further fine-tuning or quantisation, with available 4-bit GGML and GPTQ versions for CPU and GPU inference respectively.