# Playground

#### Access [Playground](https://aiplatform.console.vngcloud.vn/playground)​ <a href="#truy-cap-playground" id="truy-cap-playground"></a>

### Model Testing Guide

1. Select the model type based on your task: Chat, Completion, Image Generation, Embedding
2. CSelect a model: In the right-hand panel, choose a model from the list
3. Set up the prompt
   * In the System prompt section, define instructions or context for the model. You can also add examples to guide its behavior.
4. Adjust parameters:

<figure><img src="/files/1hJEzXoXgB5GlNv1DcH5" alt=""><figcaption></figcaption></figure>

5. Start experimenting:

* Enter your prompt and observe how different models respond.

6. Save configuration:

* Copy as code: Click the icon `<> API`
* Select programming language:

  * Curl: This command can be used in Terminal or Command Prompt, allowing you to test the API directly without writing code.
  * Python: Provides ready-to-use Python code snippets for your projects.
  * Node.js: Provides Node.js code snippets, suitable for web and server-side applications.

  <figure><img src="/files/ybHaPjSqba4O6u5HFlyd" alt=""><figcaption></figcaption></figure>

### Model Parameter Tuning Guide

When using Playground, you can adjust the following parameters to control model outputs. These parameters may vary depending on the selected model type.

**Chat Models**

* Past messages included: Number of recent messages the model will remember.\
  Range: 1 - 20 (Default: 10).
* Maximum token output: Maximum number of tokens the model can generate.\
  Range: 1 - 10000 (Default: 800).
* Temperature: Controls randomness of responses.
  * 0: Deterministic, consistent results.
  * < 1: Focused and accurate results, ideal for summarization or Q\&A.
  * ≈ 1: Encourages creativity and diversity in responses.
  * Range: 0 - 2 (Default: 1).
* Top P: Probability threshold for token selection, balancing naturalness and diversity.
  * Range: 0 - 1 (Default: 0.7).
* Presence Penalty: Penalizes repeated words, encouraging new topics and reducing repetition. hình tạo ra các chủ đề mới và giảm lặp từ.
  * Range: -2 to 2 (Default: 0).

**Completion Models**

* Maximum token output: Same as **Chat Model**
  * Range: 1 - 10000 (Default: 2048).
* Temperature: Same as **Chat Model**
  * Range: 0 - 2 (Default: 1).
* Top P: Same as **Chat Model**
  * Range: 0 - 1 (Default: 0.7).

**Image Generation Models**

* Number of Images: Number of images generated per request.
  * Range: 1 - 4 (Default: 1).
* Image size: Size of generated images.
  * Options: 256x256, 512x512, 1024x1024 (Default: 1024x1024).
* Response format: Output format of results (typically Base64).

**Embedding Models**

* Presence Penalty: Penalizes repeated words, encouraging diversity and reducing repetition.
  * Range: -2 to 2 (Default: 0).

For more detailed parameters, refer to the documentation of vLLM and [OpenAI](https://platform.openai.com/docs/api-reference/introduction).

### Model Comparison

This feature allows you to run and compare outputs from two models simultaneously.

* Click the Compare button to open two parallel chat windows.
* You can enter a prompt and send it to both models at the same time to evaluate their responses and performance.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.vngcloud.vn/vng-cloud-document/ai-stack/ai-platform/model-as-a-service/playground.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.