Closed jdeck88 closed 1 month ago
I think the model should be hosted on HuggingFace (https://huggingface.co/), which allows us to have different model versions, and has been designed for ML models. HuggingFace also allows us to generate a DOI from the model that we can use for the identifier. And we can host the code there as well. Here is an example for a model page: https://huggingface.co/HuggingFaceTB/SmolLM-1.7B . I have an account already and will work on getting the model up there for testing.
Awesome! once you get the DOI, please add this to the 300 record test test as our model_uri value.
Sounds like we agree on the need for using a DOI in model_uri value and @rdinnager is taking care of obtaining that. this particular issue can be closed.
We need to populate a field containing a unique identifier for the model that was used. Suggest a model_uri field.
robugur said: can we just the one you just created eg https://biscicol.org/api/v1/inaan/ark:/92250/machine_interpreted_model_run_1? Downstream, we may have metadata about the models we bundle up and can link to this identifier, right?
@jdeck88 said That link above is just an echo of what is contained in the string. What if we make a protocols.io DOI right now? Russell, maybe this protocol can be made under your account? Eventually, this protocol will be referenced in a publication and details/further information can be filled out later. I just made a protocols.io account and a sample protocol in about 5 minutes and got a doi for it. it takes a bit of time to for the DOI to become "activated" but at least you can get a DOI right away to refer to the model you used and update details later.
@ramonawalls said: I'm okay with this. I just wish it were recorded in a github issue.
In Sum, Russell will be creating a DOI pointing to his model protocol at protocols.io and we'll be using that.