Unleash the Power of WizardLM & Dominate ChatGPT!

Table of Contents

1. Introduction
2. What is Wizard LM?
   1. Training Technique: Evil Instruct
3. Installation Process
   1. Web UI Installation
   2. Downloading the Wizard LM Model
4. Comparing Wizard LM and Vicuna Models
   1. Summary Evaluation: TV Series Breaking Bad
   2. Article Summary Evaluation
   3. Math Test: Solving Equations
   4. Translation Test: English to French
   5. Coding Test: HTML Page with Random Background Color
5. Pros and Cons of Wizard LM Model
6. Conclusion

Wizard LM: The Ultimate Local NM Model

In this article, we will dive into the world of Wizard LM, a powerful new enemy model that is making waves in the local NM arena. With its impressive training technique known as Evil Instruct and a small yet mighty 7 billion parameters, Wizard LM is shaping up to be one of the best local NNA models currently available. In this guide, we will explore what sets Wizard LM apart, how to install it on your computer, and compare its performance to the renowned Vicuna model. So, grab your wands and spellbooks, and let's dive into the magical world of Wizard LM.

What is Wizard LM?

Wizard LM is a local NM model that stands out from the crowd due to its impressive performance and unique training technique called Evil Instruct. With a modest 7 billion parameters, Wizard LM has been designed to follow complex instructions and deliver exceptional results. Unlike traditional models that rely on human-created instructor options, Wizard LM uses LM's to automatically generate complex instructions, resulting in improved performance and even outperforming models like Chan GPT.

Training Technique: Evil Instruct

The training technique used by Wizard LM, known as Evil Instruct, is a revolutionary approach that leverages the power of LM's to automatically generate complex instructions. By making the instructions more intricate, the model's performance is enhanced, and it can follow even the most complex guidelines. This approach has proven to be successful, providing Wizard LM with an edge in the local NM space.

Installation Process

To harness the power of Wizard LM, You need to install it on your computer. The installation process involves setting up the web UI and downloading the necessary files. Follow the steps below to get started:

Web UI Installation

Before downloading Wizard LM, you need to install the Uber bugatt Tech generation web UI. If you haven't done so already, refer to the updated video guide that walks you through the installation process. Once installed, ensure that you are running the latest version of the web UI by using the provided update_windows.bot file. Then, run the start_windows.bot file to access the web UI.

Downloading the Wizard LM Model

To download the Wizard LM model, we will be using the 4-bit contest model created by a user called The Block. To simplify the process, a pre-packaged version of the model files has been created. Simply click on the link provided in the description to access the files. Copy the full name displayed on the page, go to the web UI, navigate to the model section, and paste the name in the "Load Custom Model or URL" field. Click on "Download" to initiate the download process. Once downloaded, refresh the list of models in the web UI, select Wizard LM with the appropriate parameters (128 for group size and Llama for model Type), click on "Reload the Model," and don't forget to click "Save Settings."

Comparing Wizard LM and Vicuna Models

To truly understand the capabilities of Wizard LM, we will compare its performance to the renowned Vicuna model, both of which have 7 billion parameters. Through a series of tests, we will evaluate their abilities in various scenarios.

Summary Evaluation: TV Series Breaking Bad

To gauge the summarization skills of Wizard LM and Vicuna, we will test their ability to provide accurate plot summaries of the TV series Breaking Bad. Using both models, we generate summaries and rank them using GPT4. While Wizard LM excels in providing a precise and comprehensive overview of the Show's plot, Vicuna delivers a more detailed analysis, discussing character transformations and critical acclaim. However, GPT4 rates Wizard LM's summary higher, emphasizing the need for awards and critical acclaim in Vicuna's description. The choice between the two models becomes subjective at this point, with both delivering impressive results.

Article Summary Evaluation

The next test involves generating summaries for a given article. Upon comparing the summaries produced by Wizard LM and Vicuna, GPT4 rates Wizard LM higher due to its precision and comprehensive overview of the main points. Vicuna's summary, albeit close, lacks the additional Context about other artists and broader implications of AI-generated voices in the music industry. This test result further solidifies Wizard LM's position as a powerful local NM model.

Math Test: Solving Equations

In this test, we evaluate the ability of both models to solve mathematical equations. With a simple equation presented, both Wizard LM and Vicuna attempt to solve it. Sadly, neither model succeeds in obtaining the correct solution. While this proves disappointing, it highlights the limitations of these 7 billion parameter models in handling mathematical problems.

Translation Test: English to French

Translating sentences accurately can be a challenging task. Wizard LM and Vicuna are put to the test in translating a sentence from English to French. While both models produce translations, neither achieves a Flawless result. Wizard LM's translation exhibits some inaccuracies, while Vicuna's translation fails to capture the intended meaning of the sentence. Ultimately, GPT4 rates Wizard LM higher due to its precision, albeit generously, as human judgment may yield a lower score.

Coding Test: HTML Page with Random Background Color

The coding test aims to evaluate the models' ability to generate code for an HTML page. Both Wizard LM and Vicuna are prompted to write code for a page with a button that, when pressed, changes the background to a random color. Surprisingly, Wizard LM delivers flawless code, successfully changing the background color upon button press. Conversely, Vicuna's code fails to produce the expected outcome. This impressive performance by Wizard LM showcases its potential, even with a modest 7 billion parameters.

Pros and Cons of Wizard LM Model

Pros:

• Impressive performance with 7 billion parameters
• Leveraging Evil Instruct training technique for enhanced results
• Capable of providing precise and comprehensive plot and article summaries
• Successful in generating code for specific tasks

Cons:

• Limited capabilities in solving mathematical equations
• Imperfections in translations
• Censorship limitations

Conclusion

In conclusion, Wizard LM is an exceptional local NM model that boasts impressive performance and unique capabilities. With its Evil Instruct training technique and a small yet powerful 7 billion parameters, Wizard LM has proven itself to be a strong contender in the local NNA arena. While it excels in providing accurate and comprehensive summaries, code generation, and certain translations, it falls short in mathematical problem-solving and faces censorship limitations. As the future holds the promise of even more advanced local NM models, Wizard LM sets a high standard for what can be achieved with 7 billion parameters.