Post
Nick Fisher
hydroxide.dev
did:plc:kxlkxipo2lid6pqmajvsxpxx
First time I've come across Macaron, but this looks interesting. Mixture-of-LoRA fine-tune of GLM5.1 and a custom harness for agent tasks. Lines up with the other LoRA work I looked at:
"cluster tasks that share skills and thinking patterns into one LoRA"
https://macaron.im/mindlab/research/macaron-v1-preview
[contains quote post or other embedded content]
2026-06-08T09:01:18.063Z