Google adk intentionally blocking other models

[rong-xyz]

    if llm_request.model and llm_request.model.startswith('gemini-2'):
      ...
    else:
      raise ValueError(
          f'Code execution tool is not supported for model {llm_request.model}'
      )

This is a joke right

[Jacksunwei]

This is the Gemini 2 builtin code_executor that isn't available in other models. We also have model-agnostic code executor classes. You could check out in code_executors package.

[rong-xyz]

Is this open source project's open the same as openai's open?
If other models are incapable of following the instruction/ formatting, it should be a parsing error or something else.
Not if not gemini raise error

[hangfei]

There might be some confusions on what's this code executor is.

As Jack pointed out, there are two types of code executors. One type of code executor is built into model itself. That's what Gemini-2 series support and even other variants of Gemini doesn't support. I believe the screenshot you shared is for this 1st type. The other type is general code executors that are model-agnostic, they are in this package. Even for this second type of code executors, we provide several variants to choose. Generally, we expect users to use generic code executors in this package and it's also a general industry practice.

Do you have a use case and model that you discovered doesn't work for the 2nd type of code executors? If you see some bugs or feature gaps while trying it out, we are happy to help look into them.