> For clean Markdown of any page, append `.md` to the page URL.
> For a complete documentation index, see https://docs.sarvam.ai/llms.txt.
> For full documentation content in one file, see https://docs.sarvam.ai/llms-full.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.sarvam.ai/_mcp/server.

# How to control response diversity with `top_p`

> Method used to generate text by limiting the possibilities of the next word

The `top_p` parameter controls **how much of the probability space** the model uses when selecting the next word. This is called **nucleus sampling**.

**Range:** `0` to `1`\
**Default:** `1.0`

* Lower `top_p` → model chooses from a **smaller set of highly likely words** → more focused
* Higher `top_p` → model chooses from a **broader set of words** → more diverse

---

### When to use:

| **`top_p` value** | **Behavior**                          |
| ----------------- | ------------------------------------- |
| `0.1`             | Very focused, only top 10% words used |
| `0.3`             | Controlled diversity                  |
| `0.5`             | Balanced creativity and accuracy      |
| `0.8 - 1.0`       | Very creative, open-ended responses   |
| `1.0` (default)   | Full probability space used           |

First, install the SDK:

```bash
pip install -Uqq sarvamai
```

Then use the following Python code:

```python
from sarvamai import SarvamAI

# Initialize the SarvamAI client with your API key
client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY")

# Example 1: Using default top_p (1.0): full probability space (diverse response)
response = client.chat.completions(
    model="sarvam-105b",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What is the capital of France?"}
    ],
    # top_p is not specified → uses default 1.0
)

print(response.choices[0].message.content)
```

```python
from sarvamai import SarvamAI

client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY")

# Example 2: Using top_p = 0.3: more focused, controlled response
response = client.chat.completions(
    model="sarvam-105b",
    messages=[
        {"role": "system", "content": "You are a creative storyteller."},
        {"role": "user", "content": "Tell me a story about a magical tiger."}
    ],
    top_p=0.3
)

# Receive assistant's reply as output
print(response.choices[0].message.content)
```

> Method used to generate text by limiting the possibilities of the next word