v1.0.15

What's Changed

• feat: support cache & tp for wan vace by @DefTruth in #406
• feat: support mochi-1-preview Tensor Parallelism by @gameofdimension in #408
• chore: Update README.md by @DefTruth in #409
• feat: support HunyuanDiT Tensor Parallelism by @gameofdimension in #411
• bugfix: fix summary stats from dict by @DefTruth in #412
• bugfix: fix strify error while no-cache by @DefTruth in #414
• feat: support wan vace context parallel by @DefTruth in #415
• chore: Update README.md by @DefTruth in #416
• feat: support Wan2.1-VACE Tensor Parallelism by @gameofdimension in #417
• misc: use dummy blocks for flux by default by @DefTruth in #418

Full Changelog: v1.0.14...v1.0.15

v1.0.14

What's Changed

• chore: maybe empty cache after parallelism by @DefTruth in #402
• hotfix for _attention_dispatch check by @DefTruth in #404
• fix nunchaku deps import errors by @DefTruth in #405

Full Changelog: v1.0.13...v1.0.14

v1.0.13

What's Changed

• feat: support pixart context parallel by @DefTruth in #399
• feat: support DiT-XL context parallel by @DefTruth in #400

Full Changelog: v1.0.12...v1.0.13

v1.0.12

What's Changed

• chore: update support matrix by @DefTruth in #383
• fix hyvideo cp monkey patch by @DefTruth in #384
• chore: Update README.md by @DefTruth in #385
• chore: update bench configs by @DefTruth in #386
• parallel: auto select backend if needed by @DefTruth in #387
• feat: support cogvideox context parallel by @DefTruth in #388
• feat: support Tensor Parallelism for Chroma by @gameofdimension in #389
• feat: support cogview4 & consisid cp by @DefTruth in #391
• chore: Update CP/TP docs by @DefTruth in #392
• feat: support chorma-hd cp by @DefTruth in #393
• feat: support Kandinsky5 Tensor Parallelism by @gameofdimension in #394
• feat: fully support nunchaku flux cp by @DefTruth in #396
• feat: fully support nunchaku qwen-image cp by @DefTruth in #397
• hotfix for qwen-image-lightning nunchaku + cp by @DefTruth in #398

Full Changelog: v1.0.11...v1.0.12

v1.0.11

What's Changed

• chore: update parallelism docs by @DefTruth in #346
• chore: add params modifier docs by @DefTruth in #347
• chore: Update cache-dit desc by @DefTruth in #348
• chore: Update cache-dit desc by @DefTruth in #349
• chore: Update DBCache docs by @DefTruth in #350
• refactor: move cache_factory to caching by @DefTruth in #351
• fix example utils key error by @DefTruth in #352
• feat: add wan2.2 cp example by @DefTruth in #354
• feat: add ltxvideo cp example by @DefTruth in #357
• bugfix: fix ltxvideo context parallel noise result by @DefTruth in #361
• chore: Update README.md by @DefTruth in #362
• chore: Update README.md by @DefTruth in #363
• chore: Update README.md by @DefTruth in #364
• chore: Update README.md by @DefTruth in #365
• chore: Update README.md by @DefTruth in #366
• feat: make cache compatible with block-level cp by @DefTruth in #369
• feat: use template cp for native attn by @DefTruth in #371
• feat: add Qwen-Image-Lightning CP/TP example by @DefTruth in #372
• feat: support TP for hunyuanimage2.1/video by @gameofdimension in #373
• chore: Update Tensor Parallelism docs by @DefTruth in #376
• chore: Update examples by @DefTruth in #377
• misc: add custom attn dispatch env by @DefTruth in #378
• feat: add diffusers' hyimage21 adapter by @DefTruth in #379
• feat: support hyimage21 context parallel by @DefTruth in #381
• feat: support hyvideo context parallel by @DefTruth in #382

Full Changelog: v1.0.10...v1.0.11

v1.0.10

What's Changed

• feat: mark check forward pattern default false by @DefTruth in #321
• feat: support transformer-only api by @DefTruth in #323
• chore: cache-dit recommended by HelloGithub by @DefTruth in #328
• chore: Update News by @DefTruth in #329
• feat: support FLUX.1 w/ Tensor Parallelism by @gameofdimension in #326
• feat: support longcat-video w/ 4/8-bits by @DefTruth in #325
• chore: Update News by @DefTruth in #334
• feat: support Qwen-Image tensor parallelism by @gameofdimension in #331
• chore: update badges by @DefTruth in #336
• feat: refactor tensor parallelism dispatch by @DefTruth in #337
• chore: update qwen-image tp example by @DefTruth in #338
• refactor context parallelism dispatch by @DefTruth in #340
• feat: Support Wan series tensor parallelism by @gameofdimension in #341
• docs: add quantize api docs by @DefTruth in #343
• chore: update quantize api docs by @DefTruth in #344
• docs: add parallelism docs by @DefTruth in #345

Full Changelog: v1.0.9...v1.0.10

v1.0.9

What's Changed

• chore: Update News by @DefTruth in #314
• chore: Update README.md by @DefTruth in #315
• feat: allow empty cache_config w/ parallelism by @DefTruth in #317
• feat: flux nunchaku + context parallelism by @DefTruth in #318
• chore: add perf mode to example by @DefTruth in #319

Full Changelog: v1.0.8...v1.0.9

v1.0.8

What's Changed

• deps: minumum install dependencies by @DefTruth in #312
• feat: refactor parallelism backends by @DefTruth in #313

Full Changelog: v1.0.7...v1.0.8

v1.0.7

What's Changed

• fix some typo & parallel config check by @DefTruth in #299
• remove wrong comments by @DefTruth in #301
• feat: Qwen-Image context parallelism example by @DefTruth in #303
• feat: support kandinsky5 pipeline by @DefTruth in #304
• feat: support Photoroom/PRX pipeline by @DefTruth in #308
• chore: add Kandinsky-5 cache assets by @DefTruth in #310
• chore: add Kandinsky-5 cache assets by @DefTruth in #311

Full Changelog: v1.0.6...v1.0.7

⚡️Hybird Context Parallelism

cache-dit is compatible with context parallelism. Currently, we support the use of Hybrid Cache + Context Parallelism scheme (via NATIVE_DIFFUSER parallelism backend) in cache-dit. Users can use Context Parallelism to further accelerate the speed of inference! For more details, please refer to 📚examples/parallelism.

from cache_dit import ParallelismConfig

cache_dit.enable_cache(
    pipe_or_adapter, 
    cache_config=DBCacheConfig(...),
    # Set ulysses_size > 1 to enable ulysses style context parallelism.
    parallelism_config=ParallelismConfig(ulysses_size=2),
)
# Then, run with torchrun cmd:
# torchrun --nproc_per_node=2 parallel_cache.py