Describe the feature
Currently the models adopts liblora to implement LoRa when training huge models. But the original one is not easy to train and distribute since the artifacts are huge.
Huggingface peft is a better Lora/prompt tuning library that can be easily trained and distributed because the additional parameters are stored separately and one massive model can be loaded with different peft parameters to serve different purposes.
So adding peft support will make Chat more production ready.
Describe the feature
Currently the models adopts liblora to implement LoRa when training huge models. But the original one is not easy to train and distribute since the artifacts are huge.
Huggingface peft is a better Lora/prompt tuning library that can be easily trained and distributed because the additional parameters are stored separately and one massive model can be loaded with different peft parameters to serve different purposes.
So adding peft support will make Chat more production ready.