How to assign different LoRA to different sub-agents?

[dalek-who]

Is there an existing way to specify a different LoRA configuration for each sub-agent​ during training?

Currently, the agent_match parameter in the Trainer can select which sub-agents are trained. Howerver, is it possible to config which sub-agents use which LoRAs?

import agentlightning as agl

trainer = agl.Trainer(
    ...,
    adapter={"agent_match": "a regex pattern for trained sub-agents"}
)

Related issue: #110

[ultmaster]

Do VERL support optimizing multiple LORAs simultaneously?