Oh and @sasha.bsky.social is there too. She is seemingly mostly silly 💘

I'm hoping bsky is a place that I can curate to be more silly and playful. I think I would enjoy expressing and engaging more online again. I do miss the digital space

One of the things that I miss about my old way of expressing on the internet was the silliness - the playfulness. #oldinternet #ladyengs

[sidenote: I'm tired of labelling threads as (2/n). But there is something still fun about the unit of information one can convey in a single tweet. It's breaks down a stream of consciousness into unintended bullet points I guess. Units of meaning]

BERT models are so good for so many reasons:

+ embeddings!
+ they are small and much easier to serve!
+ out of the box they perform well on specialized tasks that LLMS have to be prompted with (summarization and classification in particular)

I love this description of them here as a Honda Civic.

LOL I am in 30 min of trying to create a simple uv env + jupyter notebook to try out this model and failing, also failing using colab because it doesn't have the latest version of transformers from git

Unfortunately this is the default ML experience.

Thanks for the shout-out Paul. We focus on various dialects of African English, Zulu, Sesotho, Swahili, Hausa, Yoruba and West &Central African french.

We'll be offering West African french as part of the API by next week