#Would "MedGemma Text only model" be ever added to Openrouter?
31 messages · Page 1 of 1 (latest)
i think thered be legal issues with doing so
Why?
Isn't it open source?
medical stuff, idk a guarantee for confidentiality, providers following regulations regarding that
so if I want to provide a solution based on such AI model, or field in general, I have to self host?
since I saw a Dutch tool called Autoscriber, it uses GPT 3.5 and GPT 4
based on their peer reviewed academic research paper.
why do you want to use medgemma?
its multimodality?
because it is specialized on the medical field
I saw that it out performs many models on the out put.
For analyzing documents and images and getting medical reports.
so multimodality
Yea, but even the text only model is good.
GPT is general, not specilised.
as soon as you drop the image requirement, there are dozens of models that are better despite being general
some are western, some are chinese
some are open, some are closed
some are big, some are small
but (med)gemma is outcompeted outside of multimodal
Alright, thanks for the heads up
So it'd be better to use GPT 5 Thinking model
yeah gpt-5 (thinking) is a much better model
Is it possible to fine tune it through Openrouter? Since I think I cannot access it.
fine tuning can be expensive and openrouter doesn't expose it yet
Alright, so it would be better to do prompt engineering of the preprompt, and add a dataset filled with examples, like the reference data?
prompt engineering is worth looking into yes
models can learn in context
there are even frameworks like dspy for that kind of thing (although i've never used them)
Alright, thank you so much ❤️
gl