How to improve factual accuracy with `wiki_grounding`

If wiki_grounding is enabled, the model uses a RAG (Retrieval-Augmented Generation) approach:

It retrieves relevant chunks from Wikipedia based on the user’s question.
It then uses those chunks to ground its answer — making the response more accurate and fact-based.

What is RAG?

RAG = Retrieval-Augmented Generation.

Instead of only relying on the model’s internal knowledge (which can be outdated or incomplete), RAG allows the model to:

Look up external data sources (in this case, Wikipedia)
Condition its response on those sources

Why enable `wiki_grounding`?

When to use	Benefit
Factual Q&A	Improves accuracy
Educational content	Provides verified explanations
”What is X?” or “Explain Y” type questions	Uses up-to-date info from Wikipedia
Historical or scientific queries	Reduces hallucination

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: Using wiki_grounding to improve factual accuracy
7 response = client.chat.completions(
8     model="sarvam-105b",
9     messages=[
10         {"role": "system", "content": "You are a knowledgeable history professor."},
11         {"role": "user", "content": "Who was the first woman to win a Nobel Prize?"}
12     ],
13     wiki_grounding=True,
14     reasoning_effort="high",
15     temperature=0.3
16 )
17 
18 # Receive assistant's reply as output
19 print(response.choices[0].message.content)

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: Using wiki_grounding to improve factual accuracy
7	response = client.chat.completions(
8	model="sarvam-105b",
9	messages=[
10	{"role": "system", "content": "You are a knowledgeable history professor."},
11	{"role": "user", "content": "Who was the first woman to win a Nobel Prize?"}
12	],
13	wiki_grounding=True,
14	reasoning_effort="high",
15	temperature=0.3
16	)
17
18	# Receive assistant's reply as output
19	print(response.choices[0].message.content)