ComfyUI Node
SDXL HSWQ FP8 Quantizer (Spec-aligned)
Quantize an SDXL model to FP8 using HSWQ calibration stats.
SDXL HSWQ FP8 Quantizer (Spec-aligned)
- model
- model
◄hswq_stats_pathoutput/hswq_stats/sdxl_calib_session_01.pt►
◄keep_ratio0.25►
◄bins8192►
◄num_candidates1000►
◄refinement_iterations10►
◄scaledfalse►
◄inject_comfy_metadatatrue►
◄log_levelBasic►
CategoryQuantization
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| hswq_stats_path | STRING | output/hswq_stats/sdxl_calib_session_01.pt | — |
| keep_ratio | FLOAT | 0.250–1 | — |
| bins | INT | 8192512–65536 | — |
| num_candidates | INT | 100050–5000 | — |
| refinement_iterations | INT | 100–30 | — |
| scaled | BOOLEAN | false | — |
| inject_comfy_metadata | BOOLEAN | true | — |
| log_level | COMBO | Basic | 3 options: Basic, Verbose, Debug |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |