Qwen3 embedding models
15 points by kaycebasques
15 points by kaycebasques
These are a pretty big deal: they’re Apache 2 licensed and now occupy top spots on the well respected MTEB leaderboard: https://huggingface.co/spaces/mteb/leaderboard
Agreed. I also think that embedding models being open weight or open source is extra important, because they end up being the foundation of your database index. Much more annoying to switch embedding models than to switch LLMs.
32K of input context is also pretty attractive. They’re not transformers.js-compatible yet, right?
Someone got them working in that here: https://huggingface.co/spaces/webml-community/qwen3-embedding-webgpu
the model never finishes loading for me, doesnt seem to be any networkactivity either. is the source of the experiment available elsewhere for the webgpu port you mentioned, wasn’t able to link back?
It has to load a 560MB of data before the page can start working - it might require Chrome.
Here’s the source code: https://huggingface.co/spaces/webml-community/qwen3-embedding-webgpu/tree/main