ComfyUI Node
Checkpoint Loader Resident
Checkpoint loader with the KJ DiffusionModelLoader-style tuning knobs plus GPU-resident ingest. It reuses live checkpoint components when possible and composes missing outputs from the model, CLIP, and VAE loaders instead of materializing a broad checkpoint state dict.
Checkpoint Loader Resident
- MODEL
- CLIP
- VAE
◄ckpt_name▾►
◄weight_dtype▾►
◄compute_dtypedefault►
◄patch_cublaslinearfalse►
◄sage_attentiondisabled►
◄enable_fp16_accumulationfalse►
◄policy_override—►
CategoryGPU Resident Loader/loaders
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| ckpt_name | COMBO | Checkpoint file to load. | |
| weight_dtype | COMBO | 7 options: default, fp8_e4m3fn, fp8_e4m3fn_fast, fp8_e5m2, fp16, bf16, +1 | |
| compute_dtype | COMBO | default | Compute dtype to apply to the diffusion model patcher after load. |
| patch_cublaslinear | BOOLEAN | false | Toggle ComfyUI's cublas_ops performance feature. |
| sage_attention | COMBO | disabled | Patch optimized attention override on the loaded model. |
| enable_fp16_accumulation | BOOLEAN | false | Set torch.backends.cuda.matmul.allow_fp16_accumulation. |
| policy_overrideopt | STRING | Optional residency policy override. Connect Set Global Residency Policy here. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |
| CLIP | CLIP | — |
| VAE | VAE | — |