Normally you can tune models to return json. But, fortunately, Gemini
First install the lib
pip install google-genai
Here is a snippet i used to receive JSON response from Google GenAi.
from google import genai
prompt = '<prompt here>'
client = genai.Client(api_key='YOURKEYHERE')
chat = client.chats.create(model='gemini-2.0-flash-exp',
config={
'response_mime_type': 'application/json',
}
)
for chunk in chat.send_message_stream(prompt):
print(chunk.text)
You can also use PyDantic to customise the schema
from google.genai import types
from pydantic import BaseModel
class Profile(BaseModel):
name: str
email: str
response = client.models.generate_content(
model="gemini-2.0-flash-exp",
contents="Give me info about 2 random persons",
config=types.GenerateContentConfig(
response_mime_type="application/json",
response_schema=Profile,
),
)
print(response.text)
You can customise the schema as in the examples.
response = client.models.generate_content(
model="gemini-2.0-flash-exp",
contents="Give me information for the United States.",
config=types.GenerateContentConfig(
response_mime_type="application/json",
response_schema={
"required": [
"name",
"population",
"capital",
"continent",
"gdp",
"official_language",
"total_area_sq_mi",
],
"properties": {
"name": {"type": "STRING"},
"population": {"type": "INTEGER"},
"capital": {"type": "STRING"},
"continent": {"type": "STRING"},
"gdp": {"type": "INTEGER"},
"official_language": {"type": "STRING"},
"total_area_sq_mi": {"type": "INTEGER"},
},
"type": "OBJECT",
},
),
)
print(response.text)
Json types can be one of those:
- “TYPE_UNSPECIFIED”
- “STRING”
- “NUMBER”
- “INTEGER”
- “BOOLEAN”
- “ARRAY”
- “OBJECT”
Written by
Abdur-Rahmaan Janhangeer
Chef
Python author of 7+ years having worked for Python companies around the world
Suggested Posts
Chat with your own data/text files using chatGPT and LangChain
There are many different approaches to chatting with your own data. One way is to upload your PDF t...
Understand langChain to Quickly Get Started with Production Codebases
LangChain is a superb library to productionize LLMs. It has few concepts which make it great. If you...
Simple JSON parser without knowing anything beforehands
Table of Content Introduction Scanning the text Looking for specific characters Identifying element...
