https://www.boundaryml.com logo
Powered by
thinking_token = 0 in model config
# help
i
How do I set reasoning budget to 0 for any function, client or call?
I am using gemini-2.5-flash and I don't need reasoning capabilities. How do I turn it off?
@Vaibhav (BAML 🐑) ?
v
heya!
you just do it via client definitions
any options you add into your 
options
field are basically pass through and you can always see the 
raw curl
via the playground to confirm
let me share some tokens!
https://ai.google.dev/gemini-api/docs/thinking#set-budget given this, you would do:
Copy code
client MyThing {
  provider "google-ai"
  options {
    ...
    generationConfig {
      thinkingConfig {
        thinkingBudget 1024
      }
    }
  }
}
hope that helps!
c
I have a related question. How can I get the agent's thoughts back from BAML? I don't want to add it to the response (I think it will cost extra).
v
Do you mean access the thinking tokens? You should check out the collector in BAML
Or on-tick
c
@Vaibhav (BAML 🐑) not just the tokens. I want to get the actual thoughts as a string.
v
Yep! The collector exposes that. Via the collector you can access the raw http response from the model provider and get the thoughts from there!
PreviousNext
Powered by