The page has been translated by Gen AI.

Setting Playground Parameters

When you set the model’s parameters in the Playground, they are incorporated into the API request along with the prompt entered in the chat window, allowing you to view the model’s response results in the same way as when you actually call the API.

Model Parameter Overview

The model parameters that can be set in the Playground are as follows.

Parameter (UI display name)	API parameter name	Description	Configuration method
System Prompt (System Prompt)	`messages.role`	Instruction defining the model’s role and response format	System Prompt Parameter Settings
Maximum number of generated tokens (Max Completion Tokens)	`max_completion_tokens`	Maximum token length the model can generate	Setting the maximum generation token count parameter
Temperature(Temperature)	`temperature`	Adjust the response’s creativity/randomness	Set temperature parameters
Top‑p Sampling (Top P / Nucleus Sampling)	`top_p`	Adjust the probability distribution range of response candidate tokens	Setting Core Sampling Parameters
Prompt Safety Model(Prompt Safety Model)	`prompt_safety_model`	Whether to apply the harmful prompt detection filter	Prompt Safety Model Parameter Setting
Thinking (Chat Template Kwargs)	`chat_template_kwargs`	Whether the model’s inference (Reasoning) process is exposed	Configure Thinking Parameters

Table. Configurable model parameter types

Guide

When you set the model parameters, the new settings take effect from messages sent after the configuration.
Even if you change the parameter settings during a conversation, the previously exchanged dialogue does not reflect the new settings.

Setting system prompt parameters

System prompt parameters are instructions that specify the role, tone, and constraints the model must follow throughout the conversation. It is applied consistently throughout the entire conversation session, regardless of the individual questions the user enters in the chat window.

To set the system prompt parameters, follow the steps below.

Click the All Services > AI-ML > Simple AI Inference menu. 1. Navigate to the Service Home page of Simple AI Inference.
On the Service Home page, click the PlayGround menu. 2. Go to the PlayGround page.
Click the settings icon next to the model name on the PlayGround page. 3. Parameter Settings The popup window opens.
Parameter Settings After setting the system prompt parameters in the popup window, click the Confirm button.
Item Explanation
Default Empty
Use the model’s default behavior
Input range Enter freely within 0 ~ 1,000 characters
Requirement status Select whether the system prompt is required
Table. System Prompt Parameter Settings

Item	Explanation
Default	Empty Use the model’s default behavior
Input range	Enter freely within 0 ~ 1,000 characters
Requirement status	Select whether the system prompt is required

Notice

If the system prompt is not set, the model responds as a general-purpose assistant without any specific role instructions.
It is recommended to specify the response format (e.g., JSON output, use of a specific language) in the system prompt when you want to enforce it.

Setting the maximum generation token count parameter

The maximum generation token count parameter limits the maximum number of tokens (token) that the model can generate in a single response. To set the maximum generation token count parameter, follow these steps.

Click the All Services > AI-ML > Simple AI Inference menu. 1. Go to the Service Home page of Simple AI Inference.
On the Service Home page, click the PlayGround menu. 2. Go to the PlayGround page.
Click the settings icon next to the model name on the PlayGround page. 3. Parameter Settings The popup window opens.

Parameter Settings In the popup window, after setting the maximum generated token count parameter, click the Confirm button.

Item	Explanation
default	1,024
Input range	100 ~ 8,000 The input range varies depending on the maximum context length of the selected model It varies depending on the maximum context length of the selected model If a short answer (classification, one-line summary, etc.) is needed: set it low, around 256 ~ 512 If a brief answer (classification, one-line summary, etc.) is needed: set it low, around 256 ~ 512 If a long answer (generating long documents, code writing, etc.) is needed: it is recommended to set it to 2,048 or higher
unit	Token(Token)

Item

Explanation

default

1,024

Input range

100 ~ 8,000

The input range varies depending on the maximum context length of the selected model

It varies depending on the maximum context length of the selected model

If a short answer (classification, one-line summary, etc.) is needed: set it low, around 256 ~ 512

If a brief answer (classification, one-line summary, etc.) is needed: set it low, around 256 ~ 512

If a long answer (generating long documents, code writing, etc.) is needed: it is recommended to set it to 2,048 or higher

unit

Token(Token)

Table. Maximum generation token count parameter settings

Information

The larger the configured value, the longer responses become possible, but response time and cost increase as well.
If the response length exceeds the configured limit, the answer may be truncated in the middle, even if the sentence is not complete.

Setting the temperature parameter

The temperature parameter controls the randomness in the process of selecting the next token during response generation. To set the temperature parameter, follow these steps.

Click the All Services > AI-ML > Simple AI Inference menu. 1. Navigate to the Service Home page of Simple AI Inference.
On the Service Home page, click the PlayGround menu. 2. Navigate to the PlayGround page.
Click the settings icon next to the model name on the PlayGround page. 3. Parameter Settings The popup window opens.

Parameter Settings In the popup window, set the temperature parameter, then click the Confirm button.

Item	Explanation
default	0.7
Input range	0.0 ~ 2.0 The lower the value, the more consistent and deterministic response; the higher the value, the more diverse and creative responses are generated For fact-based Q&A, code generation, and classification: 0.0 ~ 0.3 recommended For general conversation and document summarization: 0.5 ~ 0.8 recommended For idea brainstorming and creative content: 0.9 ~ 1.2 recommended

Item

Explanation

default

0.7

Input range

0.0 ~ 2.0

The lower the value, the more consistent and deterministic response; the higher the value, the more diverse and creative responses are generated

For fact-based Q&A, code generation, and classification: 0.0 ~ 0.3 recommended

For general conversation and document summarization: 0.5 ~ 0.8 recommended

For idea brainstorming and creative content: 0.9 ~ 1.2 recommended

Table. Temperature parameter setting items

Notice

When set to 0, it generates responses that are consistently close to identical for the same input (though it does not guarantee perfect reproducibility).
The temperature parameter and the core sampling parameter can be set simultaneously, but it is generally recommended to adjust only one of them.

Setting Core Sampling Parameters

Core sampling is a method that restricts sampling of the next token to only the candidate token set whose cumulative probability meets or exceeds the configured value (P). Like the temperature parameter, it controls response diversity, but it differs in that it truncates the probability distribution itself. To set the core sampling parameters, follow the steps below.

Click the All Services > AI-ML > Simple AI Inference menu. 1. Navigate to the Service Home page of Simple AI Inference.
On the Service Home page, click the PlayGround menu. 2. Go to the PlayGround page.
On the PlayGround page, click the settings icon next to the model name. 3. Parameter Settings popup window opens.

Parameter Settings In the popup window, after setting the core sampling parameters, click the Confirm button.

Item	Explanation
default	1.0
Input range	0.0 ~ 1.0 The lower the value (e.g., 0.1), only a small number of high-probability tokens remain as candidates, resulting in a conservative response The higher the value, the more candidate tokens are sampled When set to 1.0, sampling is performed over all candidate tokens to generate a response

Item

Explanation

default

1.0

Input range

0.0 ~ 1.0

The lower the value (e.g., 0.1), only a small number of high-probability tokens remain as candidates, resulting in a conservative response

The higher the value, the more candidate tokens are sampled
- When set to 1.0, sampling is performed over all candidate tokens to generate a response

Table. Core sampling parameter setting items

Notice

If you set both the temperature parameter and the core sampling parameter low, the response may become overly repetitive or monotonous.
When the setting is set to 1.0, sampling is performed over all candidate tokens, allowing the temperature parameter setting to control diversity.

Setting Prompt Safety Model Parameters

The prompt safety model parameter is a feature that pre‑checks, using a separate safety model, whether the user‑provided prompt contains content that may be harmful or potentially violate policy (such as violence, hateful expression, requests for personal information, etc.). To set the prompt safety model parameters, follow the steps below.

All Services > AI-ML > Simple AI Inference Click the menu. 1. Navigate to the Service Home page of Simple AI Inference.
On the Service Home page, click the PlayGround menu. 2. PlayGround navigate to the page.
Click the settings icon next to the model name on the PlayGround page. 3. Parameter Settings The popup window opens.

Parameter Settings In the popup window, after setting the prompt safety model parameters, click the Confirm button.

Item	Explanation
default	Unused
Input format	Set whether to use the prompt safety model parameter When enabled: the prompt safety model first assesses the risk of the user prompt, and only forwards it to the main model if no issues are found If deemed a policy violation, it returns an advisory message without invoking the main model When disabled: the prompt is sent directly to the main model without a safety check

Item

Explanation

default

Unused

Input format

Set whether to use the prompt safety model parameter

When enabled: the prompt safety model first assesses the risk of the user prompt, and only forwards it to the main model if no issues are found
- If deemed a policy violation, it returns an advisory message without invoking the main model

When disabled: the prompt is sent directly to the main model without a safety check

Table. Prompt safety model parameter settings

Notice

Even if you do not use a prompt safety model, the model’s inherent default safety policy remains unchanged.

Setting Thinking Parameters

Before the model generates the final answer, you can enable the reasoning (Reasoning) process that analyzes the problem step by step and set whether to expose that process along with the response. Setting the thinking parameter can improve response quality in complex math problems, multi-step logical reasoning, and similar tasks.

Notice

Using the think parameter can consume additional tokens during inference, which may increase response time and cost.
The thinking parameter can be used only when a model that supports Reasoning is selected. * If you select a model that does not support reasoning, it will be disabled (grayed out) and cannot be selected.

To set the thinking parameter, follow the steps below.

All Services > AI-ML > Simple AI Inference Click the menu. 1. Navigate to the Service Home page of Simple AI Inference.
On the Service Home page, click the PlayGround menu. 2. Go to the PlayGround page.
Click the settings icon next to the model name on the PlayGround page. 3. Parameter Settings The popup window opens.

Parameter Settings After setting the thinking parameter in the popup window, click the Confirm button.

Item	Explanation
default	Depends on the model
Input format	Set whether to use the thinking parameter Enabled: The model’s inference process (thinking area) is displayed in a collapsed form at the top of the response and can be expanded by clicking Disabled: Generates only the final answer directly without the inference process Provides faster response speed for typical conversations or simple Q&A

Item

Explanation

default

Depends on the model

Input format

Set whether to use the thinking parameter

Enabled: The model’s inference process (thinking area) is displayed in a collapsed form at the top of the response and can be expanded by clicking

Disabled: Generates only the final answer directly without the inference process
- Provides faster response speed for typical conversations or simple Q&A

Table. Thinking parameter setting items

How-to Guides

References