Visit the official CodeGPT Plus website to access the Playground.
To access the Playground, you first need to create an account at Code GPT Plus.
✨ You can access CodeGPT Plus for free here.
In the Playground, you have the ability to:
Interact with Agents: Engage with the agents and fine-tune their behavior.
Configuration and Adjustment: Easily configure and adjust agent behavior through specific context documents.
External Sharing: Access an external link to share your agents with others.
Embedding: Embed your agents seamlessly into any website.
There are two ways to log in to CodeGPT Plus:
Creating a Username and Password:
Enter your Fullname.
Provide your Email account (please note that you will have access).
Create a password (if you forget it in the future, you can restore it here).
Initiating Authentication via Google or GitHub:
Simply follow the Google or GitHub authentication process.
⚠️ Remember, if you choose any of these two options, you must always log in using the same method.
In this panel, you have access to the personal data associated with your account. You can edit your name and include a phone number if you wish.
If you want to delete your account, you can use the Delete Account option.
For Free-trial accounts, personal data will be removed from our database within a 30-calendar-day interval.
For paid accounts, personal data is deleted, and the payment subscription is canceled for the remaining contracted time. After this period, any data and agents created in the playground will be lost unless the plan is renewed before the canceled plan expires.
This panel displays your subscriptions and the start date of the contracted plan. If you have a contracted plan, such as Basic, you can manage it through the "Manage Plan" button. In the opened menu on a parallel tab, you can change:
Additionally, you can access your payment history.
If you have an OpenAI or LangSmith account, you can connect it to CodeGPT Plus.
ℹ️ Note: For CodeGPT Plus users, there's no need to change the OpenAI API Key by default.
In this section, you have the option to modify the OpenAI API Key, which is necessary only if you have a non-free OpenAI account. Instructions for obtaining the API Key are available in this video tutorial.
For LangSmith API key, follow the instructions provided on their service page to obtain a LangSmith Project and LangSmith Endpoint
The Documents feature is designed to provide specific knowledge to the agents. The benefit of this feature is the ability to train an agent with specific information, thus enriching the agent's prompt. Specialization allows for improved agent performance in the desired task. You can fine-tune the model to perform optimally in specific situations and contexts, which can be challenging to achieve with more general models—all of this without the need to generate lines of code.
Upload a File
Allowed formats for uploading a file include
.txt, .csv, .xls, .xlsx, .pdf, .doc, .docx, .pages, .dot, .dotx, .log, .odt, .rtf. You can drop the file onto the panel or click the
Upload a GitHub Repository
If you have a paid plan account, you can import a GitHub repository using the
Import GitHub button. Remember that the repository must be public, and the size is limited to 10MB per upload.
Add a New Folder
You can manage files within named folders. Note that you can move files by dragging them onto a folder or by navigating into the folder and uploading there.
⚠️ Regarding File Size and Storage Capacity
Remember that depending on the contracted plan, you have a maximum file storage limit. While the maximum file upload size is 10MB, the vectorization process's resulting file size depends on various factors:
The Language Model, for example, if you're using GPT-3, GPT-4, etc. The model has its tokenization strategy, hence the number of tokens varies based on the model.
Knowing the Text Length, the number of tokens required depends on the text length. You can divide your text into tokens to estimate. Note that some models may have limits on the maximum number of tokens they can process in a single call.
If you are using the GPT-3 model from OpenAI, according to the documentation available on their website, this model would divide the following sentence into 23 tokens:
"The artificial intelligence is transforming many aspects of our daily lives. From virtual assistants to recommendation systems, AI is present in various applications."
GPT-3 uses "tokens" that are not just words but smaller segments of text.
If we assume each token occupies approximately 4 bytes in memory, we can calculate the conversion weight as follows:
23 tokens * 4 bytes/token = 92 bytes
In other words, the longer the sentence, the more bytes the vector will occupy in this case.
Error Training Document: This error often occurs with spreadsheets or Excel files due to empty rows. Be sure to clean your dataset of None rows before uploading the document.
File Not Supported: Please check the mentioned file formats before proceeding.
Error Uploading Document: Check the file size and calculate your storage availability.
Unexpected Token: This error applies to JSON files or similar code files, except if they come from a GitHub repository.
Agents interact with the environment, process information, and make decisions. In natural language, an agent might be a model responding to user queries in a "playground" setting.
In the Settings section, you can modify the following:
Name: Remember, the name is crucial for the agent's prompt.
Model: Currently, Playground offers the following models:
GPT 3.5 Turbo
GPT 4 16k
Prompt: A message or symbol indicating the system is ready to receive input. It invites users to provide commands or information, commonly seen in command-line interfaces. :::caution Avoid using curly brackets in the prompt text. :::
Link chat: You can embed this link on your webpage or any public space.
You can fine-tune the personality in this section:
Temperature: It measures the "creativity" of the agent; a higher temperature allows for more inventive responses.
⚠️Be cautious, depending on the task, agents may or may not be allowed to hallucinate.
Referents: The number of references (in chunks) provided from the Documents.
Maximum Length: The maximum number of tokens for the agent's response. Sometimes, it's important to limit the number of response tokens during interactions.
Agent ID: You can use this ID for API and SDK programming.
Functions: Improve the prompt and align the model response with functions. This is an OpenAI feature, so we recommend following this documentation.
Folders: You can delve into document folders to pair agents with specific knowledge from uploaded files in the [Documents] section.
No Folders: Also, some orphan documents can be paired with the chosen agent.
⚠️ If required documents are not available in the Agent settings, it means the documents were not trained. Go back to the Documents section and check the document status.
Document on Agent Setting Doesn't Appear: For large documents, it takes some time (minutes) to appear as trained and available in Agent settings.
Enjoying the Playground? Test the agent's responses here. Depending on your plan, the number of interactions is independent of it.
Save chats in the left panel, name each chat, or organize them into folders.
Error 400: Possible reasons include CodeGPT Plus API Key issues, often due to an expired plan. You can update your payment method or transition from a free trial to a payment plan.
Server Error: Communication with the model server is disrupted. This can occur if users change the API key server in Playground Settings, and their API Key server lacks credit or is unavailable.
Agents Not Responding: Check your prompt configuration; perhaps you added curly brackets, or the agent's name doesn't match the current one.