Crafting Conversational Chatbots With Vicuna-13B: Using Open-Source LLMs for Enhancing Dialogueby@mikeyoung44
1,028 reads
1,028 reads

Crafting Conversational Chatbots With Vicuna-13B: Using Open-Source LLMs for Enhancing Dialogue

by Mike YoungJuly 25th, 2023
Read on Terminal Reader
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

let me introduce you to Vicuna-13B - an open-source conversational AI model that can help bring your chat to life.
featured image - Crafting Conversational Chatbots With Vicuna-13B: Using Open-Source LLMs for Enhancing Dialogue
Mike Young HackerNoon profile picture

Startup founders and builders, are you looking to create magical conversational experiences for your users? Want to build chatbots that feel almost human? Well, let me introduce you to Vicuna-13B - an open-source conversational AI model that can help bring your visions to life.

In the world of startups, nailing product-market fit is everything. And one of the best ways to achieve that is by creating delightful user experiences. This is where conversational AI comes in. Chatbots and virtual assistants powered by large language models like Vicuna-13B help you:

  • Understand Users Better: Have natural conversations with users to gain insights into their needs, pain points, and motivations.

  • Guide Users: Guide users through complex tasks and workflows using an intelligent assistant. Reduce drop-offs in sign-ups, onboarding, and transactions.

  • Engage and Retain Users: Build habit-forming product experiences with messaging and notifications powered by AI.

  • Stand Out from Competition: Wow your users with human-like conversational experiences well beyond basic FAQs and scripts.

But most large language models are proprietary and inaccessible. This is where Vicuna-13B shines. Let's explore how you can leverage it to take your startup to the next level.

How Vicuna-13B Works: Fine-Tuned on Real Conversations

Vicuna-13B is built by fine-tuning the powerful LLaMA architecture on 70,000 real multi-turn conversations from ChatGPT. This conversational data helps it handle chatbot interactions naturally.

According to evaluations, Vicuna-13B reaches over 90% of ChatGPT's quality while outperforming other publicly available models. The key is its training in real dialog makes conversations contextual, natural, and human-like.

Use Cases: Build Smarter Products

Here are some ways you can infuse conversational intelligence into your product with Vicuna-13B:

  • Intelligent Chatbots: Create chatbots that can answer complex questions, recommend products, and have witty conversations.

  • Conversational Commerce: Guide users through transactions while answering questions. Reduce cart abandonment.

  • Hyper-Personalization: Use conversational data to understand user needs and offer personalized content and recommendations.

  • Automate Support: Automate common support queries to resolve customer issues faster. Focus human agents on complex tickets.

  • Create Content: Automate content creation for landing pages, blogs, emails, etc. by describing what you need.

The best part is getting started with Vicuna-13B is easy. You don't need an AI Ph.D. Just read this article to find out how to use it.

How Does it Work? Training Vicuna-13B with Real ChatGPT Conversations

Vicuna-13B is built by fine-tuning the LLaMA architecture on a dataset of approximately 70,000 multi-turn conversations collected from These conversations contain real examples of how users interact with ChatGPT, helping teach the model to converse naturally.

The training process enhances Stanford's Alpaca model to handle longer conversational contexts and optimize GPU memory usage. With improvements like gradient checkpointing, the researchers were able to train Vicuna-13B within a day on 8 A100 GPUs while keeping costs low by leveraging spot instances.

According to preliminary evaluations using GPT-4 as a judge, Vicuna-13B reaches over 90% of the quality of ChatGPT and Google's Bard model. It also outperforms other openly available models like LLaMA and Alpaca in over 90% of test cases.

This demonstrates the power of combining a strong foundation like LLaMA with real conversational data to produce an advanced open-source chatbot.

Limitations of Vicuna-13B

While Vicuna-13B is a powerful and versatile chatbot model, it does have some limitations that users should be aware of:

  • Response Accuracy: As with any language model, Vicuna-13B's responses may not always be completely accurate. It can generate plausible-sounding answers, but it might occasionally produce incorrect or nonsensical responses.

  • Context Understanding: While Vicuna-13B is capable of handling multi-turn conversations to some extent, it may still struggle with long and complex dialogues. In such cases, the model might lose context and generate less relevant responses.

  • Bias and Fairness: Like many language models, Vicuna-13B can inadvertently reflect biases present in the training data. This could lead to biased or unfair responses in certain scenarios, especially if the training data contains biased language or viewpoints.

  • Computation Resources: Vicuna-13B is a large language model, and running it requires significant computational resources, particularly for fine-tuning and serving. Users with limited computing capabilities might face challenges in deploying the model.

  • Fine-Tuning Data: Vicuna-13B's performance heavily relies on the quality and diversity of the fine-tuning data. If the dataset used for fine-tuning is limited or unrepresentative, the model's performance may suffer.

  • Ethical Use: As with any AI model, it is crucial to use Vicuna-13B ethically and responsibly. Avoid deploying the model for malicious purposes, spreading misinformation, or generating harmful content.

  • Commercial Use Restrictions: Vicuna-13B is intended for research and non-commercial use only. Commercial applications of the model may infringe upon usage restrictions and terms of service.

  • Evaluation Framework: The evaluation of chatbot performance, including Vicuna-13B, remains a complex and evolving field. Preliminary evaluations using GPT-4 are insightful but may not capture all nuances of the model's performance.

Understanding these limitations will help users make informed decisions when incorporating Vicuna-13B into their projects and applications. As AI technology continues to evolve, addressing these limitations and striving for continuous improvement will be essential in the pursuit of more powerful and responsible AI models.

Understanding the Inputs and Outputs of Vicuna-13B

Before we dive into using Vicuna-13B to solve specific problems, it's essential to understand its inputs and outputs. The model takes several inputs that influence its behavior and generates a response as its output. Here are the key inputs:

  1. Prompt: The prompt is the initial text provided to Vicuna-13B, serving as a starting point for generating a response.

  2. Max_length: This parameter limits the maximum number of tokens in the generated response.

  3. Temperature: Temperature controls the randomness of outputs. Higher values make the responses more random, while lower values make them more deterministic.

  4. Top_p: When decoding text, this parameter samples from the top p percentage of most likely tokens. Lower values filter out less likely tokens.

  5. Repetition_penalty: Penalty for repeated words in the generated text; 1 is no penalty, values greater than 1 discourage repetition, and less than 1 encourages it.

  6. Seed: Seed for a random number generator for reproducibility

  7. Debug: A boolean to provide debugging output in the logs

As for the output, Vicuna-13B generates responses in the form of a JSON array of strings. Here's what it looks like:

  "type": "array",
  "items": {
    "type": "string"
  "title": "Output",
  "x-cog-array-type": "iterator",
  "x-cog-array-display": "concatenate"

Step-by-Step Guide: Using Vicuna-13B to Enhance Your Chatbot

Now, let's explore how to leverage Vicuna-13B to enhance your chatbot.

Whether you're a developer or not, there are multiple ways to interact with the model. If you don't like coding and just want to take it for a spin, check out the model demo.

If you're ready to start coding so you can eventually implement Vicuna into your product, you can use the Replicate API. Here's a step-by-step guide:

Step 1: Install the Node.js Client and Authenticate

To begin, install the Node.js client by running npm install replicate. Next, copy your API token from your account settings on and set it as an environment variable: export REPLICATE_API_TOKEN=YOURTOKEN.

Step 2: Run the Model

With the Node.js client and your API token set up, you can now run the model using the following code:

import Replicate from "replicate";

const replicate = new Replicate({ auth: process.env.REPLICATE_API_TOKEN });

const output = await
    input: {
      prompt: "Your chatbot prompt here..."

Step 3: Set Up a Webhook (Optional)

If you want to receive notifications when the prediction is complete, you can set up a webhook URL. This can be beneficial for real-time applications.

For example:

const prediction = await replicate.predictions.create({
  version: "6282abe6a492de4145d7bb601023762212f9ddbbe78278bd6771c8b3b2f2a13b",
  input: {
    prompt: "Your chatbot prompt here..."
  webhook: "",
  webhook_events_filter: ["completed"]

By following these steps, you'll be able to interact with Vicuna-13B and witness its impressive chat powers in action.

Comparing Vicuna-13B with Similar Models: Alpaca and LLaMA 7B

Vicuna-13B, Alpaca, and LLaMA 7B are all language models that excel in various natural language processing tasks. Let's explore their similarities and differences:


Vicuna-13B is a powerful language model fine-tuned on ChatGPT interactions. It is designed to generate human-like responses given a prompt, making it suitable for chat-based conversation, question-answering, and text-based recommender systems. The model can effectively respond to user queries, hold meaningful conversations, and provide personalized recommendations based on user preferences. Check out the comparison article with LLaMA here.


Alpaca is another instruction-following model fine-tuned from the LLaMA 7B model on 52K instruction-following demonstrations. It behaves qualitatively similarly to OpenAI’s text-DaVinci-003 while being surprisingly small and easy/cheap to reproduce. Alpaca is designed to generate responses based on specific instructions provided to it, making it suitable for tasks requiring instruction-based language generation.


LLaMA (and the new LLaMA v2) is a language model implemented using Transformers. It is trained to understand and generate human-like text based on the patterns and information it learns from a large dataset. LLaMA 7B has a wide range of use cases, including language generation, text completion, and language understanding tasks. It can be used for writing articles, generating dialogue for virtual characters, predictive typing, chatbots, and language translation systems. You might like this article comparing LLaMA-v2-Chat to Alpaca.

Use Cases

  • Vicuna-13B: Vicuna excels in chat-based conversation, question answering, and text-based recommender systems. It is ideal for creating interactive and engaging chatbots and providing accurate responses to user questions.
  • Alpaca: Alpaca is specialized in instruction-following tasks. It can follow specific instructions provided to it and generate responses based on those instructions.
  • LLaMA 7B: LLaMA 7B is a versatile language model capable of various language tasks, including text generation, text completion, and language understanding. It can be applied in multiple applications, such as content generation, predictive typing, and summarization.

Technical Implementation

  • Vicuna-13B and Alpaca both use fine-tuning techniques to enhance their performance based on specific datasets and tasks.

  • LLaMA 7B is implemented using Transformers, a type of deep learning model known for its success in natural language processing tasks.

Each model may have its unique limitations, which should be considered based on the specific use case and requirements.

In summary, Vicuna-13B, Alpaca, and LLaMA 7B are powerful language models with distinct strengths and applications. While Vicuna-13B shines in chat-based conversation and question answering, Alpaca is ideal for instruction-following tasks. On the other hand, LLaMA 7B's versatility allows it to be used in a wide range of language-related applications. When choosing the most suitable model, it is essential to consider the specific use case and the type of language tasks required.

Finding Similar Models with is an invaluable resource for discovering AI models that cater to various creative needs. If you're interested in finding models similar to Vicuna-13B, you can follow these steps:

Step 1: Visit

Head over to to begin your search for similar models.

Use the search bar at the top of the page to search for models with specific keywords related to your use case. For instance, if you're looking for chatbot models similar to Vicuna-13B, you can try using keywords like "chatbot," "conversation," or "question-answering."

Step 3: Filter the Results

On the left side of the search results page, you'll find several filters that can help you narrow down the list of models. You can filter and sort the models by type (e.g., Image-to-Image, Text-to-Image), cost, popularity, or specific creators.

For instance, if you're looking for text-based chatbot models, you can filter the models by selecting the "Text-to-Text" type. Additionally, you can sort the models by popularity or cost to find the most widely-used or budget-friendly options.

Step 4: Explore the Results

Browse through the search results to find models that catch your interest. Click on each model's name to access its detailed information page, which includes capabilities, use cases, technical specifications, and more.

Step 5: Compare and Choose

Compare the features and capabilities of the models you've shortlisted. Consider factors such as response accuracy, context understanding, computational requirements, and ethical considerations. This will help you decide which model aligns best with your specific needs and use case.

Using, you can efficiently explore and compare various AI models, enabling you to make informed decisions and select the most suitable model for your creative projects.

Conclusion: Embrace the Creative Possibilities of AI

I hope this guide has inspired you to explore the creative possibilities of AI with Vicuna-13B and other incredible models. By leveraging, you can continue to discover and experiment with cutting-edge AI technologies. Don't forget to subscribe to for more tutorials, updates on new AI models, and a wealth of inspiration for your next creative project. Happy building and exploring the world of AI!

For more updates and insights, you can also follow me on Twitter and check out my build-in-public stories on Let's embark on this AI journey together!

Also published here.