Intrinsics should default to parameter values in the intrinsic's `io.yaml` file

[nrfulton]

[lastras]

As per our discussion today, which was centered on temperature, all the intrinsics owners should codify the recommended temperature for evaluating their intrinsic in io.yaml, which I assume will be "0" for most of them, for reproducibility reasons.

[nrfulton]

There are two actions:

1. Ensuring that this happens in Mellea
2. set temp=0 in io.yamls for all of the intrinsics (@lastras to put this in the release checklist for adapters)

[radum2275]

@nrfulton this means adding temp: 0 in the parameters section of the io.yaml file, right? or is it temperature? thanks!

[jakelorocco]

@radum2275, I believe it is temperature spelled out: https://github.com/generative-computing/mellea/blob/main/mellea/formatters/granite/base/util.py#L279.

[frreiss]

All parameters should be spelled the same as the OpenAI chat completions API. Note that the Hugging Face backend code path only implements a subset of these parameters.

[pronics2004]

@frreiss @lastras @nrfulton What was the consensus on the adapter name and where to add in io.yaml

cc:@inkpad

[lastras]

use exactly the same name you have been using in the past (should match the existing directory structure in HuggingFace). The location is model in io.yaml

[pronics2004]

use exactly the same name you have been using in the past (should match the existing directory structure in HuggingFace). The location is model in io.yaml

Posting @ink-pad comment on this --

This variable is reserved for model-name mapping (e.g granite-4.0-micro) so that API request for inference is done properly. I don't think we can use the same for naming. It would be something like:

name: requirement-check
model: ~

And name (or whatever we decide to call it) needs to be added to common/intrinsics/constants as an optional field

[nrfulton]

Closed by #972