A Hugging Face community post describes training LoRA adapters for a model tailored to a task and the harness used to run it. The post says training is underway and links to a live view of the losses.
Key points
The proposed adapters are tailored to tasks such as data analysis, software building and coding
The post names Hermes, Pi and OpenCode as example harnesses
Training is underway, with losses visible on a linked page
Aritra, who identifies himself as being from Hugging Face, described harness adapters in a Reddit post dated November 6, 2026. The approach trains a LoRA adapter for a model for a specific task and harness combination.
The post links to a Hugging Face blog post and a page showing the training run. It does not report completed training results or give performance measurements.