elixir-nx / bumblebee

Pre-trained Neural Network models in Axon (+ 🤗 Models integration)
Apache License 2.0
1.27k stars 90 forks source link

Add annotations to QKV layers #291

Open seanmor5 opened 7 months ago

seanmor5 commented 7 months ago

Related to https://github.com/elixir-nx/axon/issues/549

LoRA targets QKV nodes. Right now the only way to do it is by inferring based on layer name. We can provide more information by giving specific metadata (such as op_name) for which layers are QKV