ComfyUI Node
Spectrum Qwen Model Patcher
A ComfyUI node in model/optimization with 10 inputs and 1 output.
Spectrum Qwen Model Patcher
- model
- MODEL
◄warmup_steps5►
◄tail_actual_steps2►
◄history_points5►
◄chebyshev_degree3►
◄max_consecutive_forecasts1►
◄ridge_lambda0.0001►
◄cache_devicemain_device►
◄force_actual_on_controltrue►
◄debugfalse►
Categorymodel/optimization
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| warmup_steps | INT | 50–64 | Number of initial solver steps forced to run as real Qwen forwards. |
| tail_actual_steps | INT | 20–32 | Number of final solver steps forced to stay on the real path. |
| history_points | INT | 52–16 | Number of real hidden-state snapshots kept for the Chebyshev fit. |
| chebyshev_degree | INT | 31–8 | Polynomial degree used by the ridge-regularized Chebyshev forecaster. |
| max_consecutive_forecasts | INT | 10–8 | Maximum number of skipped Qwen transformer passes allowed in a row before a real refresh step is forced. |
| ridge_lambda | FLOAT | 0.00010–1 | Ridge regularization strength used during the Chebyshev coefficient solve. |
| cache_device | COMBO | main_device | Where captured Qwen hidden-state history is stored for fitting and forecasting. |
| force_actual_on_control | BOOLEAN | true | Force real forwards when ControlNet/control residuals are present. Recommended. |
| debug | BOOLEAN | false | Log per-step mode decisions and a summary at the end of the run. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |