CalamitousFelicitousness 18676996d0 Sage Attention 2 + Triton workaround Qwen-Image ...

Workaround to prevent black images generated with Qwen-Image models when Sage Attention 2 is enabled with Triton as backend on devices with compute capability 8.0 and 8.6.

Simply switches back to Cuda backend for these models only.

Proof of concept, feel free to close if this is not appropriate.

2025-11-04 23:31:14 +00:00

..

apg

…

api

controlnet fixes and validation

2025-10-30 13:36:58 -04:00

cfgzero

lint formatting

2025-07-05 17:13:31 -04:00

civitai

update optional packages

2025-10-21 11:31:12 -04:00

control

controlnet allow processor to keep aspect-ratio for override images based on i2i or t2i resolution

2025-11-04 09:37:18 -05:00

dml

add builtin framepack

2025-07-08 15:47:07 -04:00

face

linting

2025-10-29 11:40:20 -04:00

facelib

add builtin framepack

2025-07-08 15:47:07 -04:00

flash_attn_triton_amd

lint formatting

2025-07-05 17:13:31 -04:00

framepack

fix framepack

2025-10-31 11:08:27 -04:00

ggml

modularize hidream loader

2025-04-16 09:11:43 -04:00

hidiffusion

css and logging tweaks

2025-07-27 08:17:41 -04:00

intel

cleanup

2025-10-26 23:46:17 +03:00

interrogate

cleanup

2025-10-26 10:01:24 -04:00

linfusion

css and logging tweaks

2025-07-27 08:17:41 -04:00

lora

add chronoedit prequants and lora loading

2025-10-31 11:08:27 -04:00

ltx

video pipeline import exception handler

2025-10-25 12:35:27 -04:00

merging

autodetect v-pred

2025-07-28 19:41:39 -04:00

onnx_impl

upadte qwen pruning and allow hf models in subfolders

2025-10-04 15:49:20 -04:00

pag

fix guidance metadata

2025-10-09 09:25:49 -04:00

perflow

settings option to disable reference models

2025-01-23 15:19:43 -05:00

postprocess

detailer expert mode enhance parsing

2025-11-04 10:41:23 -05:00

ras

add ras-sd35 experimental

2025-02-18 18:47:42 -05:00

rife

add lbm background replace with relightining

2025-07-04 15:33:16 -04:00

schedulers

add wan asymmetric vae upscaler

2025-10-28 13:55:46 -04:00

sdnq

fix typo

2025-11-04 14:30:52 +03:00

seedvr

linting

2025-10-16 19:39:05 -04:00

taesd

Fix Qwen VAE

2025-10-11 00:55:22 +03:00

teacache

lint formatting

2025-07-05 17:13:31 -04:00

todo

cleanup

2025-01-10 08:25:00 -05:00

video_models

fix save video to folder

2025-11-03 08:13:11 -05:00

cachedit.py

allow lora specifier to select model component. see changelog/wiki for docs on how to.

2025-09-23 15:25:13 -04:00

call_queue.py

cleanup/refactor state history

2025-09-12 16:12:45 -04:00

cmd_args.py

ability to disable parts of the app

2025-11-03 19:36:55 -05:00

detailer.py

detailer support for face restorer models

2025-01-18 14:29:40 -05:00

devices_mac.py

ruff updates

2024-12-20 18:58:29 -05:00

devices.py

Sage Attention 2 + Triton workaround Qwen-Image

2025-11-04 23:31:14 +00:00

errors.py

remove dead code

2025-07-05 16:47:25 -04:00

extensions.py

refactor monolithic shared.py and separate legacy options with logging

2025-07-12 13:35:41 -04:00

extra_networks.py

major refactor: remove backend original

2025-07-05 13:16:46 -04:00

extras.py

upadte qwen pruning and allow hf models in subfolders

2025-10-04 15:49:20 -04:00

face_restoration.py

…

files_cache.py

…

generation_parameters_copypaste.py

Ensure value is string before calling .lower()

2025-11-03 21:21:39 -08:00

gr_hijack.py

major refactoring of modules

2025-07-03 09:18:38 -04:00

gr_tempdir.py

fix params.txt

2025-06-25 09:11:17 -04:00

hashes.py

cleanup/refactor state history

2025-09-12 16:12:45 -04:00

history.py

update logs styles infotext

2025-02-24 08:24:14 -05:00

images_grid.py

guard against images list and avoid pipeline switches

2025-10-28 14:38:35 -04:00

images_namegen.py

guard against images list and avoid pipeline switches

2025-10-28 14:38:35 -04:00

images_resize.py

allow resize non-pil

2025-07-12 09:44:42 -04:00

images.py

networks info details include preview image metadata

2025-11-04 10:17:52 -05:00

img2img.py

remove split-attention and add attention slicing option

2025-09-28 18:01:34 -04:00

infotext_utils.py

fix ansi controle output and add model locking

2025-07-06 12:50:09 -04:00

infotext.py

update override settings

2025-10-30 14:01:32 -04:00

ipadapter.py

offline mode for controlnet/t2i/xs/lite/ipadapter/processors

2025-10-23 09:40:53 -04:00

json_helpers.py

refactor monolithic shared.py and separate legacy options with logging

2025-07-12 13:35:41 -04:00

lama.py

styles support parsed and upparsed save and apply

2025-02-11 11:33:37 -05:00

loader.py

reference load cleanup

2025-10-25 23:25:10 -04:00

localization.py

…

masking.py

fix inpaint

2025-10-31 13:39:55 -04:00

memmon.py

update memmon

2025-04-21 20:20:04 -04:00

memstats.py

add buffers and cache info

2025-09-30 14:36:55 -04:00

mit_nunchaku.py

update nunchaku

2025-10-03 12:23:24 -04:00

model_quant.py

add basic xpu gpu monitor

2025-10-26 18:55:54 -04:00

model_te.py

fix loading custom t5

2025-08-13 16:06:25 -04:00

model_tools.py

add new optimum-quanto on-the-fly and simplify quantization loading

2025-03-16 21:45:05 -04:00

modeldata.py

fix model type detect

2025-10-31 12:31:43 -04:00

modelloader.py

cleanup hf download

2025-10-26 12:02:20 -04:00

models_hf.py

fix hf downloader

2025-10-17 13:42:28 -04:00

modelstats.py

upadte qwen pruning and allow hf models in subfolders

2025-10-04 15:49:20 -04:00

modular_guiders.py

add hunyuanimage-2.1

2025-10-24 10:29:58 -04:00

modular.py

add hunyuanimage-2.1

2025-10-24 10:29:58 -04:00

olive_script.py

…

options_handler.py

fix cwd on git version check fail

2025-11-03 16:23:41 -05:00

options.py

refactor monolithic shared.py and separate legacy options with logging

2025-07-12 13:35:41 -04:00

para_attention.py

major refactor: remove backend original

2025-07-05 13:16:46 -04:00

patches.py

…

paths_internal.py

change default hfcache folder

2025-08-13 13:04:14 -04:00

paths.py

reference models indicate if ready or requires download

2025-09-14 10:30:21 -04:00

postprocessing.py

major refactor: remove backend original

2025-07-05 13:16:46 -04:00

processing_args.py

add nvidia-chronoedit

2025-10-30 19:52:29 -04:00

processing_callbacks.py

linting

2025-10-23 11:01:20 -04:00

processing_class.py

fix inpaint

2025-10-31 13:39:55 -04:00

processing_correction.py

replace boundry with boundary

2025-07-10 14:26:16 -04:00

processing_diffusers.py

revert frame index

2025-11-02 22:25:49 +03:00

processing_helpers.py

fix image str decode and update requirements

2025-09-29 10:56:03 -04:00

processing_info.py

experiments with repa-e

2025-10-27 15:28:20 -04:00

processing_vae.py

experiments with repa-e

2025-10-27 15:28:20 -04:00

processing.py

fix samples check

2025-11-02 06:18:36 -05:00

progress.py

disabling live preview should not disable progress updates

2025-10-21 11:07:47 -04:00

prompt_parser_diffusers.py

Merge pull request #4262 from nolbert82/master

2025-10-14 09:27:40 -04:00

prompt_parser_xhinker.py

cleanup reference models

2025-08-08 14:07:33 -04:00

prompt_parser.py

lint

2025-08-02 15:28:45 -04:00

rocm_triton_windows.py

lint fixes

2025-09-04 14:56:10 -04:00

rocm.py

windows gfx120x disable MIOpen

2025-11-03 00:37:59 +09:00

safe.py

linting

2025-10-16 19:39:05 -04:00

script_callbacks.py

startup tracing

2025-01-08 09:35:49 -05:00

script_loading.py

fix ansi controle output and add model locking

2025-07-06 12:50:09 -04:00

scripts_auto_postprocessing.py

major refactoring of modules

2025-07-03 09:18:38 -04:00

scripts_manager.py

lint fixes

2025-07-16 12:59:40 -04:00

scripts_postprocessing.py

fix postprocess

2025-11-04 08:23:45 -05:00

scripts.py

lint

2025-07-31 17:33:44 -04:00

sd_checkpoint.py

improve model matching

2025-10-26 12:58:26 -04:00

sd_detect.py

add tracing to model detect

2025-11-01 19:18:31 -04:00

sd_hijack_accelerate.py

lint

2025-08-01 18:26:57 -04:00

sd_hijack_dynamic_atten.py

add typing to sdpa hijacks

2025-09-10 04:43:57 +03:00

sd_hijack_freeu.py

css and logging tweaks

2025-07-27 08:17:41 -04:00

sd_hijack_hypertile.py

css and logging tweaks

2025-07-27 08:17:41 -04:00

sd_hijack_safetensors.py

improve runai-streamer integration

2025-11-02 12:14:15 -05:00

sd_hijack_te.py

cleanup/refactor state history

2025-09-12 16:12:45 -04:00

sd_hijack_utils.py

…

sd_hijack_vae.py

add wan asymmetric vae upscaler

2025-10-28 13:55:46 -04:00

sd_hijack.py

improve extensions compatibility

2025-07-05 17:42:13 -04:00

sd_models_compile.py

Unify compile Upscaler compile command

2025-09-03 18:51:21 +03:00

sd_models_utils.py

seedvt2

2025-10-12 15:35:08 -04:00

sd_models.py

curly braces for inline wildcards

2025-10-31 18:16:06 -04:00

sd_modules.py

full ui-settings refactor

2025-03-30 15:04:17 -04:00

sd_offload.py

Add WanTransformerBlock to no split

2025-11-02 18:33:44 +03:00

sd_samplers_common.py

disabling live preview should not disable progress updates

2025-10-21 11:07:47 -04:00

sd_samplers_diffusers.py

add base and max shift parameters for flow-matching samplers

2025-10-19 14:11:41 -04:00

sd_samplers.py

video tab layout

2025-10-18 14:07:52 -04:00

sd_te_remote.py

experiment: remote-t5

2025-08-15 10:56:36 -04:00

sd_unet.py

nunchaku sdxl and sdxl-turbo support

2025-09-20 21:01:21 -04:00

sd_vae_approx.py

…

sd_vae_natten.py

…

sd_vae_ostris.py

…

sd_vae_remote.py

add smollm3, add auraflow-0.2, enable taesd for auraflox

2025-07-14 07:24:20 -04:00

sd_vae_repa.py

experiments with repa-e

2025-10-27 15:28:20 -04:00

sd_vae_stablecascade.py

styles support parsed and upparsed save and apply

2025-02-11 11:33:37 -05:00

sd_vae_taesd.py

lint

2025-11-03 19:37:31 -05:00

sd_vae.py

Add VAE scale override for ChronoEditPipeline

2025-11-01 18:53:14 +00:00

server.py

…

shared_defaults.py

Enable triton flash atten option for rocm linux too

2025-09-03 16:26:00 +03:00

shared_helpers.py

refactor monolithic shared.py and separate legacy options with logging

2025-07-12 13:35:41 -04:00

shared_items.py

add tracing to model detect

2025-11-01 19:18:31 -04:00

shared_legacy.py

fix and regenerate locales

2025-08-20 14:40:39 -04:00

shared_state.py

disabling live preview should not disable progress updates

2025-10-21 11:07:47 -04:00

shared.py

fix ui_disabled

2025-11-04 12:06:32 -05:00

styles.py

curly braces for inline wildcards

2025-10-31 18:16:06 -04:00

sub_quadratic_attention.py

…

textual_inversion.py

refactor ui_models

2025-08-02 14:18:58 -04:00

theme.py

major requirements refactor

2025-07-01 13:12:21 -04:00

timer.py

cleanup

2025-11-02 06:29:19 -05:00

token_merge.py

…

transformer_cache.py

update changelog and cleanup

2025-03-27 11:53:03 -04:00

txt2img.py

remove split-attention and add attention slicing option

2025-09-28 18:01:34 -04:00

ui_caption.py

Merge branch 'dev' into patch-2

2025-10-01 20:58:58 +01:00

ui_common.py

import sdnq on video prequant model load

2025-11-01 11:20:30 -04:00

ui_components.py

modernui configurable grid images size

2025-08-31 12:39:59 -04:00

ui_control_elements.py

controlnet allow processor to keep aspect-ratio for override images based on i2i or t2i resolution

2025-11-04 09:37:18 -05:00

ui_control_helpers.py

qwen multi-image edits

2025-10-04 18:06:50 -04:00

ui_control.py

fix guidance metadata

2025-10-09 09:25:49 -04:00

ui_docs.py

add opts models_not_to_offload

2025-09-19 11:21:54 -04:00

ui_extensions.py

add opts models_not_to_offload

2025-09-19 11:21:54 -04:00

ui_extra_networks_checkpoints.py

add quantized models to reference

2025-10-25 11:17:12 -04:00

ui_extra_networks_history.py

…

ui_extra_networks_lora.py

fix lora filter by model type

2025-09-24 12:55:11 -04:00

ui_extra_networks_styles.py

add mtime to reference models

2025-08-13 20:53:39 -04:00

ui_extra_networks_textual_inversion.py

add wildcards to networks ui

2025-08-13 20:23:35 -04:00

ui_extra_networks_vae.py

handle missing version

2025-08-25 19:39:40 -04:00

ui_extra_networks_wildcards.py

Fix wildcard parents() termination guard

2025-09-02 01:27:58 -04:00

ui_extra_networks.py

networks info details include preview image metadata

2025-11-04 10:17:52 -05:00

ui_gallery.py

add opts models_not_to_offload

2025-09-19 11:21:54 -04:00

ui_guidance.py

Change the default SDXL sampler to Euler A + Linear and set Guidence scale value to 4

2025-10-24 19:58:59 +03:00

ui_history.py

job history

2025-09-12 22:14:07 -04:00

ui_img2img.py

fix guidance metadata

2025-10-09 09:25:49 -04:00

ui_javascript.py

job history

2025-09-12 22:14:07 -04:00

ui_loadsave.py

ability to disable parts of the app

2025-11-03 19:36:55 -05:00

ui_models_load.py

1000 papercuts

2025-05-13 21:51:33 -04:00

ui_models.py

add pony-v7

2025-10-26 20:05:52 -04:00

ui_postprocessing.py

add opts models_not_to_offload

2025-09-19 11:21:54 -04:00

ui_prompt_styles.py

…

ui_sections.py

ability to disable parts of the app

2025-11-03 19:36:55 -05:00

ui_settings.py

ability to disable parts of the app

2025-11-03 19:36:55 -05:00

ui_symbols.py

restyled all toolbuttons to be modernui native

2025-08-31 15:01:50 -04:00

ui_txt2img.py

fix guidance metadata

2025-10-09 09:25:49 -04:00

ui_video_vlm.py

fix duplicate ui labels

2025-10-28 14:46:58 -04:00

ui_video.py

video tab add params.txt

2025-10-22 08:39:45 -04:00

ui.py

ability to disable parts of the app

2025-11-03 19:36:55 -05:00

update.py

…

upscaler_simple.py

add wan asymmetric vae upscaler

2025-10-28 13:55:46 -04:00

upscaler.py

fix tiny vae with batched results

2025-10-18 10:44:16 -04:00

video.py

unified video save

2025-10-05 20:17:32 -04:00

windows_hip_ffi.py

windows hip arch detection fix endless loop

2025-11-03 00:28:01 +09:00

zluda_installer.py

hijack cholesky decomposition for therock pytorch

2025-09-28 18:03:02 +09:00

zluda.py

disable miopen when unavailable

2025-11-04 12:37:12 +09:00