Document non-Diffusers FP8/NVFP4 ComfyUI export#1991
Conversation
Signed-off-by: Jingyu Xin <jingyux@nvidia.com>
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
|
Important Review skippedDraft detected. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Enterprise Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
Signed-off-by: Jingyu Xin <jingyux@nvidia.com>
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #1991 +/- ##
==========================================
- Coverage 73.83% 70.31% -3.52%
==========================================
Files 519 525 +6
Lines 57960 64384 +6424
==========================================
+ Hits 42795 45272 +2477
- Misses 15165 19112 +3947
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
Signed-off-by: Jingyu Xin <jingyux@nvidia.com>
What changed
nn.ModuleDictwithout creating a Hugging Face or Diffusers model.FP8_DEFAULT_CFGandNVFP4_DEFAULT_CFG, including the NVFP4-only padding, scale swizzling, and second scale.Why
The existing example constructs a Diffusers pipeline and uses the unified Hugging Face exporter. Custom torch-based applications may instead use a regular Python owner with several attached
nn.Modulecomponents and custom checkpoint naming. They need a pipeline-independent route to ModelOpt's low-level packing APIs plus an explicit loader-specific ComfyUI adapter.Impact
Documentation only; no runtime behavior changes.
Validation
pre-commit run --files examples/diffusers/quantization/CUSTOM-TORCH-TO-COMFYUI.md examples/diffusers/README.mdast.parse.