[PR #197] Add Triton INT8 dtype dequantization #211

New issue

Open

opened 2026-10-26 22:21:37 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 22:21:37 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/comfy-kitchen/pull/197
Author: @arcusbuilds
Created: 9/24/2026
Status: 🔄 Open

Base: main ← Head: feature/triton-int8-dequantize-dtype

📝 Commits (1)

3e66dca Add Triton INT8 dtype dequantization

📊 Changes

3 files changed (+124 additions, -1 deletions)

View changed files

📝 comfy_kitchen/backends/triton/__init__.py (+12 -0)
📝 comfy_kitchen/backends/triton/quantization.py (+68 -0)
📝 tests/test_int8.py (+44 -1)

📄 Description

Summary

add a Triton implementation of dequantize_int8_simple_dtype
support scalar, elementwise, and rowwise scales, with eager fallback for other layouts
add parity, eager-fallback, and empty-input coverage

Tests

python -m pytest tests/test_int8.py -k "triton_dequantize_int8_simple_dtype" -v (11 passed)
python -m ruff check comfy_kitchen/backends/triton/quantization.py comfy_kitchen/backends/triton/__init__.py tests/test_int8.py

_{🔄 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/197 **Author:** [@arcusbuilds](https://github.com/arcusbuilds) **Created:** 9/24/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `feature/triton-int8-dequantize-dtype` --- ### 📝 Commits (1) - [`3e66dca`](https://github.com/Comfy-Org/comfy-kitchen/commit/3e66dca4a34873b0dc0472b3462edf9c32a9ba1c) Add Triton INT8 dtype dequantization ### 📊 Changes **3 files changed** (+124 additions, -1 deletions) <details> <summary>View changed files</summary> 📝 `comfy_kitchen/backends/triton/__init__.py` (+12 -0) 📝 `comfy_kitchen/backends/triton/quantization.py` (+68 -0) 📝 `tests/test_int8.py` (+44 -1) </details> ### 📄 Description ## Summary - add a Triton implementation of `dequantize_int8_simple_dtype` - support scalar, elementwise, and rowwise scales, with eager fallback for other layouts - add parity, eager-fallback, and empty-input coverage ## Tests - `python -m pytest tests/test_int8.py -k "triton_dequantize_int8_simple_dtype" -v` (11 passed) - `python -m ruff check comfy_kitchen/backends/triton/quantization.py comfy_kitchen/backends/triton/__init__.py tests/test_int8.py` --- <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 22:21:37 +00:00

bsakaguchi referenced this issue

2026-10-26 22:21:45 +00:00

[PR #211] [MERGED] fix: decline fused DeltaNet kernels below SM75 #223