The training works (either the UI, or starting `skillopt-train` directly) when the model is configured as ```yaml model: backend: qwen_chat ``` However `skillopt-eval` run does not properly set up calling these models, so the eval-only run all just fails.
The training works (either the UI, or starting
skillopt-traindirectly) when the model is configured asHowever
skillopt-evalrun does not properly set up calling these models, so the eval-only run all just fails.