How to control response diversity with top_p

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:

$ pip install -Uqq sarvamai

Then use the following Python code:

1 from sarvamai import SarvamAI
2 
3 # Initialize the SarvamAI client with your API key
4 client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY")
5 
6 # Example 1: Using default top_p (1.0) — full probability space (diverse response)
7 response = client.chat.completions(
8     messages=[
9         {"role": "system", "content": "You are a helpful assistant."},
10         {"role": "user", "content": "What is the capital of France?"}
11     ]
12     # top_p is not specified → uses default 1.0
13 )
14 
15 print(response.choices[0].message.content)

1 from sarvamai import SarvamAI
2 
3 client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY")
4 
5 # Example 2: Using top_p = 0.3 — more focused, controlled response
6 response = client.chat.completions(
7     messages=[
8         {"role": "system", "content": "You are a creative storyteller."},
9         {"role": "user", "content": "Tell me a story about a magical tiger."}
10     ],
11     top_p=0.3
12 )
13 
14 # Receive assistant's reply as output
15 print(response.choices[0].message.content)