[PR #2] [MERGED] Add stock-ComfyUI quantization formats and model-family adapters #6

New issue

Closed

opened 2026-10-26 22:28:57 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 22:28:57 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/comfy-quants/pull/2
Author: @HK416-TYPED
Created: 6/19/2026
Status: ✅ Merged
Merged: 6/19/2026
Merged by: @alexisrolland

Base: main ← Head: main

📝 Commits (1)

a15f7d2 Add stock-ComfyUI quantization formats and model-family adapters

📊 Changes

237 files changed (+33275 additions, -2 deletions)

View changed files

➕ .gitignore (+33 -0)
📝 README.md (+202 -2)
➕ configs/anima_2b_fp8.yaml (+36 -0)
➕ configs/anima_2b_fp8_e5m2.yaml (+36 -0)
➕ configs/anima_2b_mxfp8.yaml (+36 -0)
➕ configs/anima_2b_nvfp4.yaml (+36 -0)
➕ configs/qwen_image_2512_fp8_e5m2_static.yaml (+56 -0)
➕ configs/qwen_image_2512_fp8_static.yaml (+56 -0)
➕ configs/qwen_image_2512_int8_w8a8.yaml (+48 -0)
➕ configs/qwen_image_2512_mxfp8.yaml (+56 -0)
➕ configs/qwen_image_2512_nvfp4.yaml (+56 -0)
➕ configs/qwen_image_edit_2511_fp8_e5m2_static.yaml (+66 -0)
➕ configs/qwen_image_edit_2511_fp8_static.yaml (+66 -0)
➕ configs/qwen_image_edit_2511_int4_svdquant_w4a4.yaml (+27 -0)
➕ configs/qwen_image_edit_2511_int8_w8a8.yaml (+47 -0)
➕ configs/qwen_image_edit_2511_mxfp8.yaml (+67 -0)
➕ configs/qwen_image_edit_2511_nvfp4.yaml (+67 -0)
➕ configs/qwen_image_layered_fp8_static.yaml (+56 -0)
➕ configs/qwen_image_layered_int8_w8a8.yaml (+45 -0)
➕ configs/qwen_image_layered_mxfp8.yaml (+56 -0)

...and 80 more files

📄 Description

Single squashed commit. Supersedes #1 (auto-closed by a history rewrite and not reopenable).

Contents:

Public quantization docs reorganized by format + workflow.
FP8 static (E4M3/E5M2) and INT8 W8A8 (+ConvRot) offline export paths.
MXFP8 (block-32, E8M0) and NVFP4 (FP4 E2M1, two-level) microscaling formats — stock-ComfyUI native.
Qwen-Image / Qwen-Image-Edit / Qwen-Image-Layered adapters.
Anima (cosmos_predict2 DiT) — first non-Qwen model family.

Single commit author (HK416-TYPED) resolving to a real GitHub account; CLA signing still required.

_{🔄 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-quants/pull/2 **Author:** [@HK416-TYPED](https://github.com/HK416-TYPED) **Created:** 6/19/2026 **Status:** ✅ Merged **Merged:** 6/19/2026 **Merged by:** [@alexisrolland](https://github.com/alexisrolland) **Base:** `main` ← **Head:** `main` --- ### 📝 Commits (1) - [`a15f7d2`](https://github.com/Comfy-Org/comfy-quants/commit/a15f7d24c3b2e7d004c7bf6662dd838c498a4727) Add stock-ComfyUI quantization formats and model-family adapters ### 📊 Changes **237 files changed** (+33275 additions, -2 deletions) <details> <summary>View changed files</summary> ➕ `.gitignore` (+33 -0) 📝 `README.md` (+202 -2) ➕ `configs/anima_2b_fp8.yaml` (+36 -0) ➕ `configs/anima_2b_fp8_e5m2.yaml` (+36 -0) ➕ `configs/anima_2b_mxfp8.yaml` (+36 -0) ➕ `configs/anima_2b_nvfp4.yaml` (+36 -0) ➕ `configs/qwen_image_2512_fp8_e5m2_static.yaml` (+56 -0) ➕ `configs/qwen_image_2512_fp8_static.yaml` (+56 -0) ➕ `configs/qwen_image_2512_int8_w8a8.yaml` (+48 -0) ➕ `configs/qwen_image_2512_mxfp8.yaml` (+56 -0) ➕ `configs/qwen_image_2512_nvfp4.yaml` (+56 -0) ➕ `configs/qwen_image_edit_2511_fp8_e5m2_static.yaml` (+66 -0) ➕ `configs/qwen_image_edit_2511_fp8_static.yaml` (+66 -0) ➕ `configs/qwen_image_edit_2511_int4_svdquant_w4a4.yaml` (+27 -0) ➕ `configs/qwen_image_edit_2511_int8_w8a8.yaml` (+47 -0) ➕ `configs/qwen_image_edit_2511_mxfp8.yaml` (+67 -0) ➕ `configs/qwen_image_edit_2511_nvfp4.yaml` (+67 -0) ➕ `configs/qwen_image_layered_fp8_static.yaml` (+56 -0) ➕ `configs/qwen_image_layered_int8_w8a8.yaml` (+45 -0) ➕ `configs/qwen_image_layered_mxfp8.yaml` (+56 -0) _...and 80 more files_ </details> ### 📄 Description Single squashed commit. Supersedes #1 (auto-closed by a history rewrite and not reopenable). Contents: - Public quantization docs reorganized by format + workflow. - FP8 static (E4M3/E5M2) and INT8 W8A8 (+ConvRot) offline export paths. - MXFP8 (block-32, E8M0) and NVFP4 (FP4 E2M1, two-level) microscaling formats — stock-ComfyUI native. - Qwen-Image / Qwen-Image-Edit / Qwen-Image-Layered adapters. - Anima (cosmos_predict2 DiT) — first non-Qwen model family. Single commit author (HK416-TYPED) resolving to a real GitHub account; CLA signing still required. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-26 22:28:57 +00:00