Will adapters to arbitrary torch modules be maintained in future?

[acecchini]

Hi,

From the NVidia-NeMo org git README.md, it states that NeMo repo will be breaked down "into a series of functional-focused repos to facilitate code discovery" and that it "will repurpose to focus on speech". I currently use NeMo Adapters on custom torch modules, which aren't linked to HuggingFace, for SFT and PEFT. I could not find in any of the new repos such as NeMo Megatron-Bridge or NeMo AutoModel any way to use the PEFT framework on arbitrary torch modules not being linked to HF. Hence, I wanted to know if you plan to maintain this feature in future or if it will be deprecated?

Thanks!