ComfyUI Node
HSWQ Torch Compile
torch.compile for HSWQ models. Defaults: inductor + max-autotune-no-cudagraphs (safe with USDU / Distorch). Avoid cudagraphs with multi-tile USDU.
HSWQ Torch Compile
- model
- MODEL
◄backendinductor►
◄fullgraphfalse►
◄modemax-autotune-no-cudagraphs►
◄dynamicfalse►
◄dynamo_cache_size_limit64►
◄force_parameter_static_shapestrue►
◄patch_distorch_weight_casttrue►
◄debug_compile_keysfalse►
◄disable_dynamic_vramfalse►
CategoryHSWQ/torchcompile
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| backend | COMBO | inductor | Use inductor with USDU. cudagraphs often fails on multi-tile / cudaMallocAsync pools. |
| fullgraph | BOOLEAN | false | Full graph mode. Keep False for HSWQ ConvRot / Distorch paths. |
| mode | COMBO | max-autotune-no-cudagraphs | HSWQ default avoids CUDA graphs (USDU / dynamic shapes). |
| dynamic | BOOLEAN | false | Dynamic shape tracing. Prefer off unless tile shapes vary every step. |
| dynamo_cache_size_limit | INT | 640–1024 | torch._dynamo.config.cache_size_limit |
| force_parameter_static_shapes | BOOLEAN | true | torch._dynamo.config.force_parameter_static_shapes — helps avoid symbolic weight Size errors. |
| patch_distorch_weight_cast | BOOLEAN | true | Mark comfy.ops cast_bias_weight etc. as eager (Distorch / dynamic VRAM). |
| debug_compile_keys | BOOLEAN | false | Log compile keys |
| disable_dynamic_vramopt | BOOLEAN | false | Clone model with disable_dynamic=True when ComfyUI supports it. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |