CalamitousFelicitousness d75abce642 feat(attention): sparse attention settings and router stage ...

Sparse attention is a stage over the chain rather than a member of it:
one switch, and the router hands the selection to whichever active
backend advertises that it consumes a block mask, currently flex. A
backend declares that through a capability set, so the quantized kernel
joins later without touching the router.

The stage gates on the component role, self attention, a minimum
sequence length defaulting to the measured 8192 token crossover, and the
absence of a token mask or causal flag, which flex cannot combine with a
block only mask. Budgets follow a precomputed per step schedule with at
most two distinct values. Enabling the feature with no capable backend
in the chain warns and leaves attention dense rather than doing nothing
quietly.

The modular pre-forward hook now receives kwargs and publishes whatever
token layout the pipeline passes by name, so a packed sequence gets its
conditioning pinned without any model specific code. Without a layout
the whole sequence is sparsified and that is logged once per length.

2026-08-28 12:40:22 +01:00

..

apg

fix(pipelines): restore clip-skip text encoding for transformers 5.6+

2026-06-25 02:25:33 +01:00

api

add heif/heic support

2026-08-26 10:24:17 +02:00

attention

feat(attention): sparse attention settings and router stage

2026-08-28 12:40:22 +01:00

ben2

full codespell coverage

2026-06-04 12:36:10 +02:00

caption

strict typechecking

2026-08-16 12:45:50 +02:00

cfgzero

strict ruff without exclude

2026-05-11 08:36:49 +02:00

civitai

refactor(model): simplify probe cache handling

2026-07-11 20:26:12 +01:00

control

prepare queue mgmt

2026-08-22 11:24:41 +02:00

detailer

fix missing export

2026-08-22 11:38:51 +02:00

face

pipeline init reordering

2026-08-16 12:45:50 +02:00

flash_attn_triton_amd

full codespell coverage

2026-06-04 12:36:10 +02:00

framepack

placeholder video upscale

2026-08-21 10:26:17 +02:00

ggml

full codespell coverage

2026-06-04 12:36:10 +02:00

hidiffusion

full codespell coverage

2026-06-04 12:36:10 +02:00

image

add heif/heic support

2026-08-26 10:24:17 +02:00

intel

Update IPEX and ROCm to Torch 2.13

2026-07-15 04:05:26 +03:00

linfusion

cleanup logger

2026-02-19 11:09:13 +01:00

lora

perf(lora): store hosted factors at the delta's effective rank

2026-08-26 23:48:35 +01:00

ltx

docs(ltx): correct the refine default comment and stale file references

2026-08-25 23:21:21 +01:00

merging

linting

2026-07-03 22:05:33 +02:00

minimax

switch to pillow-heif

2026-08-26 07:49:26 +02:00

onnx_impl

remove directml

2026-08-10 13:43:45 +02:00

openai

add experimental openai interface

2026-07-02 14:36:44 +02:00

pag

fix(pipelines): restore clip-skip text encoding for transformers 5.6+

2026-06-25 02:25:33 +01:00

postprocess

update upscaler workflow

2026-08-22 20:10:02 +02:00

ras

strict typechecking

2026-08-16 12:45:50 +02:00

rembg

rembg add lucida

2026-07-30 14:39:30 +02:00

res4lyf

update sampler behavior and user definable fallback

2026-06-08 17:00:00 +02:00

rife

improve memory reporting, explicit interpolation and minimax tracking

2026-08-25 09:11:46 +02:00

schedulers

strict typechecking

2026-08-16 12:45:50 +02:00

seedvr

video processing preserve audio

2026-07-27 14:23:14 +02:00

sharpfin

update ruff and lint

2026-07-24 18:20:41 +02:00

taesd

strict typechecking

2026-08-16 12:45:50 +02:00

teacache

strict typechecking

2026-08-16 12:45:50 +02:00

todo

full codespell coverage

2026-06-04 12:36:10 +02:00

vae

fix(vae): honor an explicitly requested taesd variant

2026-08-18 00:54:30 +01:00

video_models

modular pipelines intercepts and profiling

2026-08-28 13:32:14 +02:00

__init__.py

refactor imports

2026-02-19 15:28:24 +01:00

cachedit.py

cleanup logger

2026-02-19 11:09:13 +01:00

call_queue.py

fix(queue): stop the queue lock from suppressing exceptions

2026-08-26 03:36:15 +01:00

cmd_args.py

remove directml

2026-08-10 13:43:45 +02:00

devices_mac.py

mps install torchsde

2026-06-20 08:15:35 +02:00

devices.py

refactor(attention): backend registry and a single sdpa router

2026-08-23 08:20:45 +01:00

enso.py

run enso extension installer after clone and update

2026-03-20 21:43:44 +00:00

errorlimiter.py

Typing cleanup

2026-05-17 10:41:09 -07:00

errors.py

profiling updates

2026-07-26 19:35:56 +02:00

extensions.py

register submodule as package

2026-08-10 12:06:33 +02:00

extra_networks.py

fix(lora): apply te networks before encode and honor lora_apply_te

2026-07-08 03:06:04 +01:00

extras.py

fix ui/misc: None desc crash, ratio/tile/lora split guards, parse_metadtaa typo, moondream type guard

2026-06-12 13:00:50 -04:00

files_cache.py

strict typechecking

2026-08-16 12:45:50 +02:00

generation_parameters_copypaste.py

Fix assorted small typos: dead __str__, unformatted messages, preview min-height, tooltip

2026-06-11 01:42:29 -04:00

gr_hijack.py

gradio temp files guard

2026-05-16 07:19:52 +02:00

gr_tempdir.py

add heif/heic support

2026-08-26 10:24:17 +02:00

hashes.py

validate reference, lint and update diffusers

2026-04-15 10:10:05 +02:00

history.py

remove process preview

2026-04-29 07:59:43 +02:00

images.py

add heif/heic support

2026-08-26 10:24:17 +02:00

img2img.py

batch processing run register after script

2026-08-23 11:34:34 +02:00

infotext_utils.py

ruff lint

2026-02-19 11:13:44 +01:00

infotext.py

fix(sampler): record and restore Default instead of the None sentinel

2026-08-10 02:44:56 +01:00

ipadapter.py

refactor detailer

2026-07-04 11:17:56 +02:00

json_helpers.py

model metadata handle invalid keys

2026-06-29 11:23:29 +02:00

lama.py

unified logger

2026-02-19 09:46:42 +01:00

loader.py

video loader use generic methods and auth

2026-08-16 09:02:40 +02:00

localization.py

cleanup logger

2026-02-19 11:09:13 +01:00

logger.py

fix file logger traceback

2026-08-23 09:59:38 +02:00

masking.py

fix rembg dependencies

2026-07-21 08:20:55 +02:00

memmon.py

torch reset dynamo cache on model reload

2026-08-04 10:58:32 +02:00

memstats.py

improve memory reporting, explicit interpolation and minimax tracking

2026-08-25 09:11:46 +02:00

migrate.py

unified logger

2026-02-19 09:46:42 +01:00

mit_nunchaku.py

strict typechecking

2026-08-16 12:45:50 +02:00

model_probe.py

feat(model): detect header metadata and nvfp4 in the model probe

2026-07-12 03:44:54 +01:00

model_quant.py

Add Lloyd-Max quantization with use_codebook

2026-08-10 22:51:32 +03:00

model_te.py

fix model/vae: wrong quant_config var, getattr guards, remote-vae latent_copy shadowing + response handling

2026-06-12 13:00:39 -04:00

model_tools.py

strict typechecking

2026-08-16 12:45:50 +02:00

modeldata.py

feat(model): register minimax h3 as a text2image base model

2026-08-10 22:11:57 +01:00

modelloader.py

video loader use generic methods and auth

2026-08-16 09:02:40 +02:00

models_hf.py

fix(hub): install the token header hijack at startup

2026-08-14 03:39:01 +01:00

modelstats.py

add storage analyzer

2026-08-02 11:22:12 +02:00

modular_guiders.py

cleanup logger

2026-02-19 11:09:13 +01:00

modular_load.py

modular pipelines intercepts and profiling

2026-08-28 13:32:14 +02:00

modular.py

cleanup logger

2026-02-19 11:09:13 +01:00

olive_script.py

modernize typing

2026-02-19 09:15:37 +01:00

options_handler.py

chore(settings): remove the bmm attention methods

2026-08-24 19:59:41 +01:00

options.py

unified logger

2026-02-19 09:46:42 +01:00

para_attention.py

cleanup logger

2026-02-19 11:09:13 +01:00

patches.py

add wildcards to image metadata and note separate template from prompt field

2026-05-19 09:45:20 +02:00

paths_internal.py

ruff lint

2026-02-19 11:13:44 +01:00

paths.py

add offload state_dict

2026-08-17 08:52:46 +02:00

platform_linux.py

strict typechecking

2026-08-16 12:45:50 +02:00

platform.py

optimize sdnq quant-on-load and add platform stats

2026-06-07 11:43:46 +02:00

postprocessing.py

add lint full script

2026-08-16 12:45:50 +02:00

processing_args.py

feat(attention): generation context

2026-08-23 08:20:46 +01:00

processing_callbacks.py

feat(attention): generation context

2026-08-23 08:20:46 +01:00

processing_class.py

fix: dead hasattr-on-dict path, double video-vae normalization, undefined self.device, hyimage if/elif, attention-mask attr typo, wrong face logged

2026-06-12 13:00:57 -04:00

processing_correction.py

fallback unpack-latents, log client auth, update login page

2026-08-19 18:03:39 +02:00

processing_diffusers.py

modular pipelines intercepts and profiling

2026-08-28 13:32:14 +02:00

processing_grading.py

fix lut

2026-05-07 07:20:20 +02:00

processing_helpers.py

feat(samplers): group sampler dropdown into labeled sections

2026-06-14 01:41:05 +01:00

processing_info.py

modular pipelines intercepts and profiling

2026-08-28 13:32:14 +02:00

processing_prompt.py

group offloading changes

2026-08-14 15:59:51 +02:00

processing_vae.py

error handle vae decode

2026-08-24 10:26:28 +02:00

processing_video.py

improve memory reporting, explicit interpolation and minimax tracking

2026-08-25 09:11:46 +02:00

processing.py

modular pipelines intercepts and profiling

2026-08-28 13:32:14 +02:00

progress.py

prepare queue mgmt

2026-08-22 11:24:41 +02:00

prompt_parser_diffusers.py

strict typechecking

2026-08-16 12:45:50 +02:00

prompt_parser_xhinker.py

fix(prompt): restore textual inversion on clip-skip>=2

2026-06-25 02:25:33 +01:00

prompt_parser.py

fix processing: init_images hasattr typo, empty-list guards, dead timer profile branch

2026-06-12 13:00:31 -04:00

rocm_triton_windows.py

refactor imports

2026-02-19 15:28:24 +01:00

rocm.py

Pin stable ROCM 7.14, rework multi arch installation

2026-08-26 01:17:54 -06:00

safe.py

strict typechecking

2026-08-16 12:45:50 +02:00

script_callbacks.py

Move functions to match new execution order

2026-04-27 17:15:05 -07:00

script_loading.py

cleanup logging

2026-06-02 08:03:27 +02:00

scripts_auto_postprocessing.py

modernize typing

2026-02-19 09:15:37 +01:00

scripts_manager.py

fix(scripts): range-check the selectable script arg slice

2026-08-18 00:36:05 +01:00

scripts_postprocessing.py

add process video

2026-07-20 21:44:43 +02:00

scripts.py

video: support for scripts/extensions

2026-07-28 10:43:58 +02:00

sd_checkpoint.py

model metadata handle invalid keys

2026-06-29 11:23:29 +02:00

sd_detect.py

Merge pull request #5017 from vladmandic/feat/minimax-h3

2026-08-11 13:34:36 +02:00

sd_hijack_accelerate.py

strict typechecking

2026-08-16 12:45:50 +02:00

sd_hijack_compile.py

refactor modular

2026-08-12 14:24:09 +02:00

sd_hijack_dynamic_atten.py

chore(settings): remove the bmm attention methods

2026-08-24 19:59:41 +01:00

sd_hijack_freeu.py

remove directml

2026-08-10 13:43:45 +02:00

sd_hijack_hfhub.py

refactor modular

2026-08-12 14:24:09 +02:00

sd_hijack_hypertile.py

chore(settings): remove dead attention options

2026-08-23 08:21:28 +01:00

sd_hijack_modular.py

feat(attention): sparse attention settings and router stage

2026-08-28 12:40:22 +01:00

sd_hijack_safetensors.py

unified logger

2026-02-19 09:46:42 +01:00

sd_hijack_schedulers.py

add ideogram prequant and set default schedule

2026-06-08 17:20:12 +02:00

sd_hijack_te.py

feat(attention): generation context

2026-08-23 08:20:46 +01:00

sd_hijack_transformers.py

complete refactor javascript to typescript and reorg frontend files and folders

2026-05-20 18:08:10 +02:00

sd_hijack_triton.py

update kernel autotune reporting

2026-08-23 14:08:21 +02:00

sd_hijack_utils.py

modernize typing

2026-02-19 09:15:37 +01:00

sd_hijack_vae.py

feat(attention): generation context

2026-08-23 08:20:46 +01:00

sd_hijack.py

initial pyright lint

2026-02-21 09:32:36 +01:00

sd_models_compile.py

refactor modular

2026-08-12 14:24:09 +02:00

sd_models_utils.py

add storage analyzer

2026-08-02 11:22:12 +02:00

sd_models.py

fix(processing): re-apply attention when the overrides change

2026-08-23 08:20:46 +01:00

sd_modules.py

…

sd_offload_aux.py

detailer.next

2026-08-09 15:05:00 +02:00

sd_offload_balanced.py

refactor offloading

2026-08-16 12:45:50 +02:00

sd_offload_group.py

reduce group offload logging

2026-08-16 12:45:50 +02:00

sd_offload_state.py

refactor offloading

2026-08-16 12:45:50 +02:00

sd_offload_utils.py

refactor offloading

2026-08-16 12:45:50 +02:00

sd_offload.py

refactor offloading

2026-08-16 12:45:50 +02:00

sd_samplers_common.py

prepare queue mgmt

2026-08-22 11:24:41 +02:00

sd_samplers_diffusers.py

handle torchsde

2026-08-22 16:08:58 +02:00

sd_samplers.py

feat(samplers): group sampler dropdown into labeled sections

2026-06-14 01:41:05 +01:00

sd_te_remote.py

RUF013 updates and typing update

2026-03-24 05:19:51 -07:00

sd_unet.py

feat(model): secondary unet override slot and ideogram4 native loading

2026-07-11 06:40:07 +01:00

sd_vae.py

feat(video): add minimax h3 joint video and audio model

2026-08-10 22:11:57 +01:00

server.py

switch to internally managed uvicorn

2026-08-06 13:24:48 +02:00

shared_defaults.py

image metadata handle correct image index

2026-08-14 10:44:56 +02:00

shared_helpers.py

unified logger

2026-02-19 09:46:42 +01:00

shared_items.py

chore(settings): remove the bmm attention methods

2026-08-24 19:59:41 +01:00

shared_legacy.py

update offloading codepaths for non-balanced

2026-08-11 22:09:12 +02:00

shared_state.py

separate progress monitoring from live preview, live preview improvements, progress details

2026-08-07 11:41:40 +02:00

shared.py

remove directml

2026-08-10 13:43:45 +02:00

storage.py

add offload state_dict

2026-08-17 08:52:46 +02:00

styles.py

group offloading changes

2026-08-14 15:59:51 +02:00

sub_quadratic_attention.py

modernize typing

2026-02-19 09:15:37 +01:00

textual_inversion.py

fix(embeddings): restore multi-vector textual inversion expansion

2026-06-25 02:25:33 +01:00

theme.py

add legacy notice

2026-06-20 21:18:35 +02:00

timer_sdnq.py

Move SDNQ to upstream SDNQ repo

2026-08-10 12:13:10 +03:00

timer.py

modular pipelines intercepts and profiling

2026-08-28 13:32:14 +02:00

token_merge.py

cleanup logging and update requirements

2026-03-13 13:02:41 +01:00

transformer_cache.py

cleanup

2026-07-21 16:04:10 +02:00

txt2img.py

detailer.next

2026-08-09 15:05:00 +02:00

ui_caption.py

css updates

2026-07-13 15:52:55 +02:00

ui_common.py

separate progress monitoring from live preview, live preview improvements, progress details

2026-08-07 11:41:40 +02:00

ui_components.py

full codespell coverage

2026-06-04 12:36:10 +02:00

ui_control_elements.py

fix(control): remove inert DWPose (Legacy) pose-model selector

2026-06-22 10:28:59 -04:00

ui_control_helpers.py

update masking

2026-05-19 11:56:20 +02:00

ui_control.py

prepare queue mgmt

2026-08-22 11:24:41 +02:00

ui_definitions.py

feat(attention): sparse attention settings and router stage

2026-08-28 12:40:22 +01:00

ui_docs.py

strict typechecking

2026-08-16 12:45:50 +02:00

ui_extensions.py

handle invalid extension

2026-07-01 18:59:20 +02:00

ui_extra_networks_checkpoints.py

minimax ui

2026-08-13 15:33:47 +02:00

ui_extra_networks_history.py

networks avoid slow yield

2026-08-01 09:10:15 +02:00

ui_extra_networks_lora.py

feat(krea2): add Krea 2 (K2) image model support

2026-06-23 04:41:54 +01:00

ui_extra_networks_styles.py

networks avoid slow yield

2026-08-01 09:10:15 +02:00

ui_extra_networks_textual_inversion.py

cleanup references to p.sd_model

2026-02-19 20:48:42 +01:00

ui_extra_networks_unet.py

networks avoid slow yield

2026-08-01 09:10:15 +02:00

ui_extra_networks_vae.py

networks avoid slow yield

2026-08-01 09:10:15 +02:00

ui_extra_networks_wildcards.py

cleanup

2026-07-27 21:30:40 +02:00

ui_extra_networks.py

add heif/heic support

2026-08-26 10:24:17 +02:00

ui_gallery.py

ui fixes

2026-05-21 09:23:10 +02:00

ui_guidance.py

- all guidance values set to -1 to enable using model defaults

2026-05-16 11:06:23 +02:00

ui_history.py

show_progress requires "full", "minimal", or "hidden"

2025-11-28 02:27:19 -08:00

ui_img2img.py

detailer.next

2026-08-09 15:05:00 +02:00

ui_javascript.py

update handlers for all authenticated workflows

2026-08-09 22:47:17 +02:00

ui_loadsave.py

refactor imports

2026-02-19 15:28:24 +01:00

ui_models_load.py

strict typechecking

2026-08-16 12:45:50 +02:00

ui_models.py

download model index ondemand

2026-08-11 12:55:59 +02:00

ui_postprocessing.py

process button busy tracking

2026-07-22 09:39:55 +02:00

ui_prompt_styles.py

…

ui_sections.py

minimax ui

2026-08-13 15:33:47 +02:00

ui_settings.py

strict typechecking

2026-08-16 12:45:50 +02:00

ui_storage.py

add storage analyzer

2026-08-02 11:22:12 +02:00

ui_symbols.py

gallery add thumbnail size slider

2026-05-04 09:03:50 +02:00

ui_txt2img.py

detailer.next

2026-08-09 15:05:00 +02:00

ui_video_vlm.py

video: support for scripts/extensions

2026-07-28 10:43:58 +02:00

ui_video.py

placeholder video upscale

2026-08-21 10:26:17 +02:00

ui.py

networks load/download model track progress

2026-05-08 14:43:39 +02:00

update.py

Fix url formatting in update module

2025-11-08 14:54:34 -08:00

upscaler_algo.py

prepare queue mgmt

2026-08-22 11:24:41 +02:00

upscaler_nvvfx.py

prepare queue mgmt

2026-08-22 11:24:41 +02:00

upscaler_simple.py

prepare queue mgmt

2026-08-22 11:24:41 +02:00

upscaler_spandrel.py

update upscaler workflow

2026-08-22 20:10:02 +02:00

upscaler_vae.py

update upscaler workflow

2026-08-22 20:10:02 +02:00

upscaler.py

update upscaler workflow

2026-08-22 20:10:02 +02:00

video.py

process add video info and metadata

2026-07-19 14:37:20 +02:00

zluda_installer.py

Revert "Avoid using args as a variable name"

2026-03-05 20:17:44 -08:00

zluda.py

unified logger

2026-02-19 09:46:42 +01:00