[PR #189] Add CUDA MXFP8 dequantization #207

New issue

Open

opened 2026-10-26 22:21:35 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 22:21:35 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/comfy-kitchen/pull/189
Author: @arcusbuilds
Created: 9/21/2026
Status: 🔄 Open

Base: main ← Head: feature/cuda-dequantize-mxfp8

📝 Commits (7)

7b4d388 WIP: add CUDA MXFP8 dequantization kernel
ea577c6 WIP: add CUDA MXFP8 dequantization launcher
9eeaed3 WIP: declare CUDA MXFP8 dequantization launcher
50f0f02 WIP: bind CUDA MXFP8 dequantization
0470a7a add CUDA MXFP8 dequantization function and constraints
196ff53 WIP: expose CUDA MXFP8 dequantization
a315437 test: cover CUDA MXFP8 dequantization

📊 Changes

4 files changed (+193 additions, -2 deletions)

View changed files

📝 comfy_kitchen/backends/cuda/__init__.py (+43 -1)
📝 comfy_kitchen/backends/cuda/dlpack_bindings.cpp (+43 -0)
📝 comfy_kitchen/backends/cuda/ops/quantize_mxfp8.cu (+77 -1)
📝 tests/test_qdq.py (+30 -0)

📄 Description

Adds CUDA support for MXFP8 dequantization.

Fixes #190

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/Comfy-Org/comfy-kitchen/pull/189 **Author:** [@arcusbuilds](https://github.com/arcusbuilds) **Created:** 9/21/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `feature/cuda-dequantize-mxfp8` --- ### 📝 Commits (7) - [`7b4d388`](https://github.com/Comfy-Org/comfy-kitchen/commit/7b4d3886b1efeb48e467d3d95ad75e0e1573ee3e) WIP: add CUDA MXFP8 dequantization kernel - [`ea577c6`](https://github.com/Comfy-Org/comfy-kitchen/commit/ea577c66895315e403e0e280f86781fb76c549cf) WIP: add CUDA MXFP8 dequantization launcher - [`9eeaed3`](https://github.com/Comfy-Org/comfy-kitchen/commit/9eeaed39ce02b03ff0ac4850534422d487fee805) WIP: declare CUDA MXFP8 dequantization launcher - [`50f0f02`](https://github.com/Comfy-Org/comfy-kitchen/commit/50f0f0283f953d21b710ea7ef2bc66cb00329d0d) WIP: bind CUDA MXFP8 dequantization - [`0470a7a`](https://github.com/Comfy-Org/comfy-kitchen/commit/0470a7a24c39cb0e6ded61c89649a37181bb661d) add CUDA MXFP8 dequantization function and constraints - [`196ff53`](https://github.com/Comfy-Org/comfy-kitchen/commit/196ff532ddef26c829b8f81a10e44b6b4da04f1b) WIP: expose CUDA MXFP8 dequantization - [`a315437`](https://github.com/Comfy-Org/comfy-kitchen/commit/a31543755b9df3bcfb5609ed0a1cc904cfda5b86) test: cover CUDA MXFP8 dequantization ### 📊 Changes **4 files changed** (+193 additions, -2 deletions) <details> <summary>View changed files</summary> 📝 `comfy_kitchen/backends/cuda/__init__.py` (+43 -1) 📝 `comfy_kitchen/backends/cuda/dlpack_bindings.cpp` (+43 -0) 📝 `comfy_kitchen/backends/cuda/ops/quantize_mxfp8.cu` (+77 -1) 📝 `tests/test_qdq.py` (+30 -0) </details> ### 📄 Description Adds CUDA support for MXFP8 dequantization. Fixes #190 --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi added the

pull-request

label

2026-10-26 22:21:35 +00:00

bsakaguchi referenced this issue

2026-10-26 22:21:44 +00:00

[PR #207] [MERGED] Optimize int8 attention with mask. #222

bsakaguchi referenced this issue

2026-10-26 22:22:14 +00:00

[PR #256] cuda: mark attn_mask .none() on sage_sdpa and sage_sdpa_prequantized for nanobind < 2.11 #263