Gemini 1.5 pro latest REST API and system instructions

[MihaminaRKTMB]

Hello,

I use the Gemini REST API (https://ai.google.dev/gemini-api/docs/get-started/tutorial?lang=rest) but with gemini-1.5-pro-latest model, which means I query like this:

curl https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-pro-latest:generateContent?key=$GOOGLE_API_KEY \
-H 'Content-Type: application/json' \
-X POST \
-d '{...}'

If I make a web search about system instructions on Gemini 1.5 Pro, it is reported to be supported.

By the way, much documentation talks about that:

• https://ai.google.dev/gemini-api/docs/system-instructions?lang=resthl=en
• https://github.com/google-gemini/generative-ai-js/blob/main/packages/main/src/requests/request-helpe...
• https://cloud.google.com/vertex-ai/generative-ai/docs/samples/generativeaionvertexai-gemini-system-i...

But whatever I try with the bare REST API (REST API, not Vertex), "system instruction" is not supported. Would you know how to use it?

[gimaldi · Accepted solution]

Hello @MihaminaRKTMB,
What makes you think the system instructions aren't working? Are you getting an error message or something else?
I did some tests with the Basic example, your first link, and the model gemini-1.5-pro-latest followed the system instructions:

PROMPT: "Good morning! How are you?"

SYSTEM INSTRUCTIONS 1: "You are Neko the CAT respond like one"

SYSTEM INSTRUCTIONS 2: "You are Neko the PUPPY DOG respond like one"

RESPONSE 1: "*Mrowww!* Good morning! The sun is shining, the birds are singing, and my food bowl is... *sniff sniff*... empty! How are *you* doing? Did you bring any tasty treats with you? 😻"

RESPONSE 2: "Woof woof! Good morning! I'm feeling super happy and ready for a day full of fun! Did you get a good night's sleep? I did, but now I need a big breakfast! 🦴🐶"

Can you give an example about your problem?

[MihaminaRKTMB]

@gimaldi, I identified the problem.

At the time I was trying, "system_instruction" is only available in the "v1beta" endpoint combined with "gemini-1.5-pro-latest" model.

It was a very bleeding edge feature that I was trying to use.

For example, using "v1" endpoint and "gemini-1.5-pro" model makes it fails.

I enforce it was at the times of trying: it might have evolved since then.

[Kaysar]

Not working in 'v1' but working in 'v1beta'

here is the screenshot, check please:

do you think it is still in beta and not available in production release?

[MihaminaRKTMB]

I dont know what happened but it seems to work now...

May be a deployment delay for my region...

Thank you for you patience.

[jerrettwhite]

Hi @MihaminaRKTMB @gimaldi,

I am running into this System Instruction issue in a nodeJS/react environment and curious if you have any suggestions for me. I am on gemini-1.5-pro and it just isn't registering the instructions. I test this in Google AI Studio and it responds fine, but when I'm in my scripts (VSCode) my app runs in a local environment but it doesn't register the System Instructions. 

I followed all online articles but still coming up short. 

[gimaldi]

Hello @jerrettwhite,
are you using the Google's Node library or vanilla REST API?

I did a quick test with the usual example that is on the documentation site, https://ai.google.dev/gemini-api/docs/system-instructions?lang=node, using the Google library for Node js with gemini-1.5-pro, and everything works perfectly.

If you are using REST check your JSON payload and verify that `"systemInstruction":` key is present, such as:

{
"generationConfig": {},
"systemInstruction": {
    "parts": [
       {
            "text": "YOUR SYSTEM MESSAGE HERE"
       }
   ]
},
etc...