Imported from https://hf.co/sarvamai/sarvam-30b.

Chat/thinking

The model supports no thinking, but it uses a different mechanism than what ollama uses so it’s currently not possible to disable thinking.

$ ollama run frob/sarvam
>>> hello
Thinking...
The user has simply said "hello". This is a very basic greeting. I should 
respond in a warm, friendly manner in the same language/script. The user 
wrote "hello" in English, so I should respond in English.

I don't need to mention anything about my purpose or limitations unless 
asked. I should just give a friendly, simple greeting response.
...done thinking.

Hello! How can I help you today? 😊

>>> /clear
Cleared session context
>>> /set nothink
Set 'nothink' mode.
>>> hello
Thinking...
The user just said "hello" in English. This is a very simple greeting. I 
should respond in a friendly, warm manner that matches the tone.

Since the user wrote in English, I should respond in English. There's no 
need to decline anything or offer alternatives here - this is just a 
standard greeting.

I should keep my response brief and friendly, and maybe ask how I can help 
them today to be useful while maintaining the warm, helpful spirit of my 
core values.

I don't need to mention anything about being an AI or my guidelines - just 
give a natural, welcoming response.
...done thinking.

Hello! How can I help you today? I'm here to assist with any questions or 
tasks you might have. 😊

Tools

$ ollama launch opencode \
    --model frob/sarvam \
    -- \
    run 'create a file called README with the contents "hello world"'

> build · frob/sarvam:latest

← Write README
Wrote file successfully.

README created with "hello world" contents.

$ cat README 
hello world