-
Notifications
You must be signed in to change notification settings - Fork 5.9k
fix bug for torch.uint1-7 not support in torch<2.6 #10368
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
torch.uint7, | ||
) | ||
|
||
if version.parse(torch.__version__) >= version.parse('2.6'): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
oh thanks so much!
can we use is_torch_version
?
diffusers/src/diffusers/utils/import_utils.py
Line 727 in c1e7fd5
def is_torch_version(operation: str, version: str): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you! Apologies that I missed this 😅
I think this should be something like:
if version <= 2.5:
# do not use any of the uintx
else:
# use the uintx types as well
I believe the uintx data types were added in 2.5 and not 2.6
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Also maybe good to have another branch that adds float8_e4m3fn
and float8_e5m2
only if torch version is >= 2.2
. int8
came in 2.0
I believe but since minimum diffusers requirement is torch>=1.4
, another branch for it might be needed. LMK if I can help with any changes
are you able to run |
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
hi @baymax591 closing this one since it is merged |
…ersion 0.32.1 Aryan (2): Fix TorchAO related bugs; revert device_map changes (#10371) Release: v0.32.1 Sayak Paul (1): fix test pypi installation in the release workflow (#10360) YiYi Xu (1): make style for huggingface/diffusers#10368 (#10370)
…ersion 0.33.0 39th president of the United States, probably (1): Fix for multi-GPU WAN inference (#10997) Abhipsha Das (1): [Model Card] standardize advanced diffusion training sdxl lora (#7615) Ahmed Belgacem (1): Fix redundant prev_output_channel assignment in UNet2DModel (#10945) Alan Ponnachan (1): Add torch_xla support to pipeline_aura_flow.py (#10365) Alexey Zolotenkov (1): Fix incorrect seed initialization when args.seed is 0 (#10964) Ameer Azam (1): Regarding the RunwayML path for V1.5 did change to stable-diffusion-v1-5/[stable-diffusion-v1-5/ stable-diffusion-inpainting] (#10476) Andreas Jörg (1): [examples/controlnet/train_controlnet_sd3.py] Fixes #11050 - Cast prompt_embeds and pooled_prompt_embeds to weight_dtype to prevent dtype mismatch (#11051) Anton Obukhov (1): Marigold Update: v1-1 models, Intrinsic Image Decomposition pipeline, documentation (#10884) Aryan (35): Fix TorchAO related bugs; revert device_map changes (#10371) Update Flux docstrings (#10423) Update variable names correctly in docs (#10435) Fix hunyuan video attention mask dim (#10454) Fix style (#10478) [LoRA] Support original format loras for HunyuanVideo (#10376) Add `_no_split_modules` to some models (#10308) Fix Latte output_type (#10558) Test sequential cpu offload for torchao quantization (#10506) Fix offload tests for CogVideoX and CogView3 (#10547) [core] Layerwise Upcasting (#10347) Improve TorchAO error message (#10627) [core] Pyramid Attention Broadcast (#9562) Refactor gradient checkpointing (#10611) Refactor OmniGen (#10771) Disable PEFT input autocast when using fp8 layerwise casting (#10685) Update FlowMatch docstrings to mention correct output classes (#10788) Refactor CogVideoX transformer forward (#10789) Module Group Offloading (#10503) Remove print statements (#10836) Some consistency-related fixes for HunyuanVideo (#10835) SkyReels Hunyuan T2V & I2V (#10837) [docs] Add CogVideoX Schedulers (#10885) [refactor] SD3 docs & remove additional code (#10882) [refactor] Remove additional Flux code (#10881) [LoRA] Support Wan (#10943) Hunyuan I2V (#10983) [LoRA] CogView4 (#10981) LTX 0.9.5 (#10968) Group offloading improvements (#11094) Fix Group offloading behaviour when using streams (#11097) [core] FasterCache (#10163) New HunyuanVideo-I2V (#11066) Improve information about group offloading and layerwise casting (#11101) Raise warning and round down if Wan num_frames is not 4k + 1 (#11167) AstraliteHeart (2): Add AuraFlow GGUF support (#10463) Add missing `isinstance` for arg checks in GGUFParameter (#10834) Bagheera (1): fix for #7365, prevent pipelines from overriding provided prompt embeds (#7926) Basile Lewandowski (1): Change KolorsPipeline LoRA Loader to StableDiffusion (#11198) Benjamin Bossan (1): [LoRA] Implement hot-swapping of LoRA (#9453) Bhavay Malhotra (1): [train_controlnet.py] Fix the LR schedulers when num_train_epochs is passed in a distributed training env (#8461) Bruno Magalhaes (1): remove unnecessary call to `F.pad` (#10620) Bubbliiiing (1): Add EasyAnimateV5.1 text-to-video, image-to-video, control-to-video generation model (#10626) C (3): [Docs] Add documentation about using ParaAttention to optimize FLUX and HunyuanVideo (#10544) Fix Graph Breaks When Compiling CogView4 (#10959) Fix Wan I2V Quality (#11087) Cheng Jin (1): Resolve stride mismatch in UNet's ResNet to support Torch DDP (#11098) CyberVy (7): Fix Callback Tensor Inputs of the SDXL Controlnet Inpaint and Img2img Pipelines are missing "controlnet_image". (#10880) Fix Callback Tensor Inputs of the SD Controlnet Pipelines are missing some elements. (#10907) Improve load_ip_adapter RAM Usage (#10948) Fix the missing parentheses when calling is_torchao_available in quantization_config.py. (#10961) Fix Flux Controlnet Pipeline _callback_tensor_inputs Missing Some Elements (#10974) Fix missing **kwargs in lora_pipeline.py (#11011) fix _callback_tensor_inputs of sd controlnet inpaint pipeline missing some elements (#11073) Célina (1): use style bot GH Action from `huggingface_hub` (#10970) Daniel Hipke (1): Add a `disable_mmap` option to the `from_single_file` loader to improve load performance on network mounts (#10305) Daniel Regado (9): IP-Adapter support for `StableDiffusion3ControlNetPipeline` (#10363) Added IP-Adapter for `StableDiffusion3ControlNetInpaintingPipeline` (#10561) IP-Adapter for `StableDiffusion3InpaintPipeline` (#10581) IP-Adapter for `StableDiffusion3Img2ImgPipeline` (#10589) [Docs] Update SD3 ip_adapter model_id to diffusers checkpoint (#10597) `MultiControlNetUnionModel` on SDXL (#10747) SD3 IP-Adapter runtime checkpoint conversion (#10718) Comprehensive type checking for `from_pretrained` kwargs (#10758) Multi IP-Adapter for Flux pipelines (#10867) Dev Rajput (1): Add correct number of channels when resuming from checkpoint for Flux Control LoRa training (#10422) Dhruv Nair (28): [CI] Add minimal testing for legacy Torch versions (#10479) [CI] Torch Min Version Test Fix (#10491) [Single File] Fix loading Flux Dev finetunes with Comfy Prefix (#10545) [CI] Update HF Token in Fast GPU Tests (#10568) [CI] Update HF Token on Fast GPU Model Tests (#10570) [CI] Update HF_TOKEN in all workflows (#10613) [CI] Fix Truffle Hog failure (#10769) [Single File] Add Single File support for Lumina Image 2.0 Transformer (#10781) [CI] Fix incorrectly named test module for Hunyuan DiT (#10854) [CI] Update always test Pipelines list in Pipeline fetcher (#10856) [Docs] Fix toctree sorting (#10894) [CI] Improvements to conditional GPU PR tests (#10859) [CI] Fix Fast GPU tests on PR (#10912) [CI] Fix for failing IP Adapter test in Fast GPU PR tests (#10915) [CI] Update Stylebot Permissions (#10931) [Single File] Add user agent to SF download requests. (#10979) [Single File] Add single file support for Wan T2V/I2V (#10991) Fix for fetching variants only (#10646) [Quantization] Add Quanto backend (#10756) [Quantization] Allow loading TorchAO serialized Tensor objects with torch>=2.6 (#11018) [Refactor] Clean up import utils boilerplate (#11026) Provide option to reduce CPU RAM usage in Group Offload (#11106) [Quantization] dtype fix for GGUF + fix BnB tests (#11159) [Docs] Update Wan Docs with memory optimizations (#11089) [WIP] Add Wan Video2Video (#11053) Add CacheMixin to Wan and LTX Transformers (#11187) Fix Single File loading for LTX VAE (#11200) Update Ruff to latest Version (#10919) Dimitri Barbot (2): Fix pipeline dtype unexpected change when using SDXL reference community pipelines in float16 mode (#10670) Fix deterministic issue when getting pipeline dtype and device (#10696) Doug J (1): Update train_text_to_image_sdxl.py (#8830) Edna (1): Add Wan with STG as a community pipeline (#11184) Eliseu Silva (7): Make passing the IP Adapter mask to the attention mechanism optional (#10346) feat: new community mixture_tiling_sdxl pipeline for SDXL (#10759) fix: [Community pipeline] Fix flattened elements on image (#10774) feat: add Mixture-of-Diffusers ControlNet Tile upscaler Pipeline for SDXL (#10951) fix: mixture tiling sdxl pipeline - adjust gerating time_ids & embeddings (#11012) fix: for checking mandatory and optional pipeline components (#11189) feat: [Community Pipeline] - FaithDiff Stable Diffusion XL Pipeline (#11188) Fanli Lin (8): [tests] fix `AssertionError: Torch not compiled with CUDA enabled` (#10356) [tests] make tests device-agnostic (part 3) (#10437) make tensors contiguous before passing to safetensors (#10761) [tests] make tests device-agnostic (part 4) (#10508) [tests] enable bnb tests on xpu (#11001) [tests] make cuda only tests device-agnostic (#11058) [tests] no hard-coded cuda (#11186) [tests] HunyuanDiTControlNetPipeline inference precision issue on XPU (#11197) G.O.D (1): fix bug for ascend npu (#10429) Giuseppe Catalano (1): Revert RePaint scheduler 'fix' (#10644) Hanch Han (1): [fix] refer use_framewise_encoding on AutoencoderKLHunyuanVideo._encode (#10600) Haoyun Qin (1): fix: support transformer models' `generation_config` in pipeline (#10779) Ikpreet S Babra (1): Fixed grammar in "write_own_pipeline" readme (#10706) Ilya Drobyshevskiy (1): fix flux controlnet bug (#11152) Inigo Goiri (1): Add support to pass image embeddings to the WAN I2V pipeline. (#11175) Ishan Modi (1): [Single File] Add single file loading for SANA Transformer (#10947) Jacob Helwig (1): Add sigmoid scheduler in `scheduling_ddpm.py` docs (#10648) Juan Acevedo (3): implementing flux on TPUs with ptxla (#10515) reverts accidental change that removes attn_mask in attn. Improves fl… (#11065) update readme instructions. (#11096) Jun Yeop Na (2): [train_dreambooth_lora.py] Fix the LR Schedulers when `num_train_epochs` is passed in a distributed training env (#10973) [doc] Fix Korean Controlnet Train doc (#11141) Junsong Chen (8): [Sana] 1k PE bug fixed (#10431) [Sana][bug fix]change clean_caption from True to False. (#10481) [Sana 4K] (#10493) [Sana] add Sana to auto-text2image-pipeline; (#10538) [Sana-4K] (#10537) [fix bug] PixArt inference_steps=1 (#11079) [fix SANA-Sprint] (#11142) add a timestep scale for sana-sprint teacher model (#11150) Junyu Chen (2): [DC-AE] support tiling for DC-AE (#10510) [DC-AE, SANA] fix SanaMultiscaleLinearAttention apply_quadratic_attention bf16 (#10595) Kenneth Gerald Hamilton (1): Fixed requests.get function call by adding timeout parameter. (#11156) Kinam Kim (1): Add STG to community pipelines (#10960) Le Zhuo (1): Add support for lumina2 (#10642) Leo Jiang (4): [Sana 4K] Add vae tiling option to avoid OOM (#10583) NPU adaption for RMSNorm (#10534) NPU Adaption for Sanna (#10409) [bugfix] NPU Adaption for Sana (#10724) Linoy Tsaban (4): small readme changes for advanced training examples (#10473) [flux lora training] fix t5 training bug (#10845) [Wan LoRAs] make T2V LoRAs compatible with Wan I2V (#11107) [Flux LoRA] fix issues in flux lora scripts (#11111) LittleNyima (1): Add CogVideoX DDIM Inversion to Community Pipelines (#10956) Lucain (1): Remove cache migration script (#10619) Luchao Qi (1): [Typo] Update md files (#10404) Marc Sun (5): Fix compatibility with pipeline when loading model with device_map on single gpu (#10390) [FEAT] DDUF format (#10037) [FEAT] Model loading refactor (#10604) store activation cls instead of function (#10832) remove format check for safetensors file (#10864) Mark (1): [Docs] Fix environment variables in `installation.md` (#11179) Marlon May (1): Add community pipeline for semantic guidance for FLUX (#10610) Mathias Parger (1): speedup hunyuan encoder causal mask generation (#10764) Max Podkorytov (1): Fix enable memory efficient attention on ROCm (#10564) Mikko Tukiainen (1): Add missing MochiEncoder3D.gradient_checkpointing attribute (#11146) Muyang Li (2): Use `randn_tensor` to replace `torch.randn` (#10535) Remove the FP32 Wrapper when evaluating (#10617) Nicolas (1): Fix train_text_to_image.py --help (#10711) Omar Awile (1): Fix documentation for FluxPipeline (#10563) Parag Ekbote (7): Notebooks for Community Scripts-5 (#10499) Fix Documentation about Image-to-Image Pipeline (#10704) Notebooks for Community Scripts-6 (#10713) Extend Support for callback_on_step_end for AuraFlow and LuminaText2Img Pipelines (#10746) Notebooks for Community Scripts-7 (#10846) Add Example of IPAdapterScaleCutoffCallback to Docs (#10934) Notebooks for Community Scripts-8 (#11128) Pierre Chapuis (1): fix default values of Flux guidance_scale in docstrings (#10982) Rahul Raman (1): Refactor instructpix2pix lora to support peft (#10205) Raul Ciotescu (1): width and height are mixed-up (#10629) SahilCarterr (8): [Add] torch_xla support to pipeline_sana.py (#10364) [Fix] Broken links in hunyuan docs (#10402) [Fix] broken links in docs (#10434) [FIX] check_inputs function in Auraflow Pipeline (#10678) [Fix] Type Hint in from_pretrained() to Ensure Correct Type Inference (#10714) [FIX] check_inputs function in lumina2 (#10784) [Fix] Docs overview.md (#10858) [Fix] fp16 unscaling in train_dreambooth_lora_sdxl (#10889) Sayak Paul (75): [chore] post release 0.32.0 (#10361) fix test pypi installation in the release workflow (#10360) [training] fix: registration of out_channels in the control flux scripts. (#10367) [LoRA] feat: support `unload_lora_weights()` for Flux Control. (#10206) [training] add ds support to lora sd3. (#10378) [LTX-Video] fix attribute adjustment for ltx. (#10426) [Tests] add slow and nightly markers to sd3 lora integation. (#10458) [LoRA] fix: lora unloading when using expanded Flux LoRAs. (#10397) [Training] QoL improvements in the Flux Control training scripts (#10461) [chore] remove prints from tests. (#10505) [LoRA] clean up `load_lora_into_text_encoder()` and `fuse_lora()` copied from (#10495) [LoRA] allow big CUDA tests to run properly for LoRA (and others) (#9845) [CI] Match remaining assertions from big runner (#10521) [Tests] skip tests properly with `unittest.skip()` (#10527) [Docs] Add negative prompt docs to FluxPipeline (#10531) [Flux] Improve true cfg condition (#10539) [LoRA] improve failure handling for peft. (#10551) [Docs] Update hunyuan_video.md to rectify the checkpoint id (#10524) [LoRA] feat: support loading loras into 4bit quantized Flux models. (#10578) [Tests] add: test to check 8bit bnb quantized models work with lora loading. (#10576) [Chore] fix vae annotation in mochi pipeline (#10585) [training] set rest of the blocks with `requires_grad` False. (#10607) [chore] change licensing to 2025 from 2024. (#10615) [Tests] modify the test slices for the failing flax test (#10630) [docs] fix image path in para attention docs (#10632) [chore] add a script to extract loras from full fine-tuned models (#10631) [Tests] conditionally check `fp8_e4m3_bf16_max_memory < fp8_e4m3_fp32_max_memory` (#10669) [tests] update llamatokenizer in hunyuanvideo tests (#10681) [bitsandbytes] Simplify bnb int8 dequant (#10401) [LoRA] fix peft state dict parsing (#10532) [Tests] Test layerwise casting with training (#10765) [chore] update notes generation spaces (#10592) [LoRA] improve lora support for flux. (#10810) [docs] add missing entries to the lora docs. (#10819) [LoRA] make `set_adapters()` robust on silent failures. (#9618) [misc] feat: introduce a style bot. (#10274) [tests] use proper gemma class and config in lumina2 tests. (#10828) [LoRA] add LoRA support to Lumina2 and fine-tuning script (#10818) [Utils] add utilities for checking if certain utilities are properly documented (#7763) [tests] test `encode_prompt()` in isolation (#10438) [CI] install accelerate transformers from `main` (#10289) [CI] run fast gpu tests conditionally on pull requests. (#10310) fix: run tests from a pr workflow. (#9696) [chore] template for remote vae. (#10849) fix remote vae template (#10852) [LoRA] restrict certain keys to be checked for peft config update. (#10808) [chore] correct qk norm list. (#10876) [Tests] fix: lumina2 lora fuse_nan test (#10911) [style bot] improve security for the stylebot. (#10908) [chore] fix-copies to flux pipelines (#10941) [Tests] Remove more encode prompts tests (#10942) Update evaluation.md (#10938) [LoRA] feat: support non-diffusers lumina2 LoRAs. (#10909) [tests] fix tests for save load components (#10977) [CI] remove synchornized. (#10980) [LoRA] remove full key prefix from peft. (#11004) [LoRA] Improve copied from comments in the LoRA loader classes (#10995) [LoRA] Improve warning messages when LoRA loading becomes a no-op (#10187) [Tests] improve quantization tests by additionally measuring the inference memory savings (#11021) [LoRA] support wan i2v loras from the world. (#11025) [LoRA] change to warning from info when notifying the users about a LoRA no-op (#11044) [Tests] restrict memory tests for quanto for certain schemes. (#11052) [LoRA] feat: support non-diffusers wan t2v loras. (#11059) [Tests] add requires peft decorator. (#11037) [CI] pin transformers version for benchmarking. (#11067) make PR GPU tests conditioned on styling. (#11099) [CI] uninstall deps properly from pr gpu tests. (#11102) [feat] implement `record_stream` when using CUDA streams during group offloading (#11081) [bistandbytes] improve replacement warnings for bnb (#11132) minor update to sana sprint docs. (#11236) [docs] minor updates to dtype map docs. (#11237) [LoRA] support more comyui loras for Flux 🚨 (#10985) fix: SD3 ControlNet validation so that it runs on a A100. (#11238) fix timeout constant (#11252) fix consisid imports (#11254) Shenghai Yuan (1): [core] ConsisID (#10140) Shitao Xiao (1): Add OmniGen (#10148) Steven Liu (9): [docs] Quantization tip (#10249) [docs] Video generation update (#10272) [docs] Fix internal links (#10418) [docs] Fix missing parameters in docstrings (#10419) [docs] uv installation (#10622) [docs] LoRA support (#10844) [docs] Update prompt weighting docs (#10843) [docs] Flux group offload (#10847) [docs] MPS update (#11212) Suprhimp (1): [feat]Add strength in flux_fill pipeline (denoising strength for fluxfill) (#10603) Teriks (5): RFInversionFluxPipeline, small fix for enable_model_cpu_offload & enable_sequential_cpu_offload compatibility (#10480) allow passing hf_token to load_textual_inversion (#10546) SDXL ControlNet Union pipelines, make control_image argument immutible (#10663) support StableDiffusionAdapterPipeline.from_single_file (#10552) Fix SD2.X clip single file load projection_dim (#10770) Thanh Le (2): Fix inconsistent random transform in instruct pix2pix (#10698) Faster set_adapters (#10777) Tolga Cangöz (3): [`Research Project`] Add AnyText: Multilingual Visual Text Generation And Editing (#8998) Update README and example code for AnyText usage (#11028) [LTX0.9.5] Refactor `LTXConditionPipeline` for text-only conditioning (#11174) Vedat Baday (2): fix(hunyuan-video): typo in height and width input check (#10684) feat(training-utils): support device and dtype params in compute_density_for_timestep_sampling (#10699) Vinh H. Pham (1): Implement framewise encoding/decoding in LTX Video VAE (#10488) Vladimir Mandic (1): flux: make scheduler config params optional (#10384) Wenhao Sun (1): Add pipeline_stable_diffusion_xl_attentive_eraser (#10579) Yaniv Galron (3): removing redundant requires_grad = False (#10628) typo fix (#10802) making ```formatted_images``` initialization compact (#10801) Yao Matrix (4): map BACKEND_RESET_MAX_MEMORY_ALLOCATED to reset_peak_memory_stats on XPU (#11191) enable 1 case on XPU (#11219) introduce compute arch specific expectations and fix test_sd3_img2img_inference failure (#11227) fix FluxReduxSlowTests::test_flux_redux_inference case failure on XPU (#11245) YiYi Xu (7): make style for huggingface/diffusers#10368 (#10370) fix offload gpu tests etc (#10366) follow-up refactor on lumina2 (#10776) [Alibaba Wan Team] continue on #10921 Wan2.1 (#10922) update check_input for cogview4 (#10966) remove F.rms_norm for now (#11126) add sana-sprint (#11074) Yih-Dar (1): Security fix (#10905) Yuqian Hong (2): create a script to train autoencoderkl (#10605) [BUG] Fix Autoencoderkl train script (#11113) Yuxuan Zhang (5): CogView4 (supports different length c and uc) (#10649) Update pipeline_cogview4.py (#10944) [Docs] CogView4 comment fix (#10957) CogView4 Control Block (#10809) Modify the implementation of retrieve_timesteps in CogView4-Control. (#11125) Zehuan Huang (1): Support pass kwargs to cogvideox custom attention processor (#10456) ZhengKai91 (1): Fix aclnnRepeatInterleaveIntWithDim error on NPU for get_1d_rotary_pos_embed (#10820) a120092009 (1): [Quantization] support pass MappingType for TorchAoConfig (#10927) alex choi (1): ensure dtype match between diffused latents and vae weights (#8391) andreabosisio (1): Typo fix in the table number of a referenced paper (#10528) baymax591 (1): bugfix for npu not support float64 (#10123) chaowenguo (4): add pythor_xla support for render a video (#10443) Update rerender_a_video.py fix dtype error (#10451) add callable object to convert frame into control_frame to reduce cpu memory usage. (#10501) add the xm.mark_step for the first denosing loop (#10530) co63oc (1): Fix pipeline_flux_controlnet.py (#11095) célina (1): Update Style Bot workflow (#11202) dependabot[bot] (2): Bump jinja2 from 3.1.4 to 3.1.5 in /examples/research_projects/realfill (#10377) Bump jinja2 from 3.1.5 to 3.1.6 in /examples/research_projects/realfill (#10984) fancydaddy (1): add from_single_file to animatediff (#10924) geronimi73 (1): AutoModel instead of AutoModelForCausalLM (#10507) hlky (59): Default values in SD3 pipelines when submodules are not loaded (#10393) Fix AutoPipeline `from_pipe` where source pipeline is missing target pipeline's optional components (#10400) Add torch_xla and from_single_file support to TextToVideoZeroPipeline (#10445) `lora_bias` PEFT version check in `unet.load_attn_procs` (#10474) LEditsPP - examples, check height/width, add tiling/slicing (#10471) Add torch_xla and from_single_file to instruct-pix2pix (#10444) Use Pipelines without scheduler (#10439) Fix HunyuanVideo produces NaN on PyTorch<2.5 (#10482) Use pipelines without vae (#10441) Update tokenizers in `pr_test_peft_backend` (#10132) Fix tokenizers install from main in LoRA tests (#10494) UNet2DModel mid_block_type (#10469) PyTorch/XLA support (#10498) Use Pipelines without unet (#10440) Fix StableDiffusionInstructPix2PixPipelineSingleFileSlowTests (#10557) Fix train_dreambooth_lora_sd3_miniature (#10554) Fix Nightly AudioLDM2PipelineFastTests (#10556) Fix batch > 1 in HunyuanVideo (#10548) Move buffers to device (#10523) Scheduling fixes on MPS (#10549) Add IP-Adapter example to Flux docs (#10633) ControlNet Union controlnet_conditioning_scale for multiple control inputs (#10666) [training] Convert to ImageFolder script (#10664) Add provider_options to OnnxRuntimeModel (#10661) Quantized Flux with IP-Adapter (#10728) EDMEulerScheduler accept sigmas, add final_sigmas_type (#10734) Add `Self` type hint to `ModelMixin`'s `from_pretrained` (#10742) Fix `use_lu_lambdas` and `use_karras_sigmas` with `beta_schedule=squaredcos_cap_v2` in `DPMSolverMultistepScheduler` (#10740) DiffusionPipeline mixin `to`+FromOriginalModelMixin/FromSingleFileMixin `from_single_file` type hint (#10811) `device_map` in `load_model_dict_into_meta` (#10851) Fix `torch_dtype` in Kolors text encoder with `transformers` v4.49 (#10816) Add SD3 ControlNet to AutoPipeline (#10888) Experimental per control type scale for ControlNet Union (#10723) Support IPAdapter for more Flux pipelines (#10708) Add `remote_decode` to `remote_utils` (#10898) Update VAE Decode endpoints (#10939) Add VAE Decode endpoint slow test (#10946) Fix loading OneTrainer Flux LoRA (#10978) Wan VAE move scaling to pipeline (#10998) Fix SD3 IPAdapter feature extractor (#11027) Use `output_size` in `repeat_interleave` (#11030) [hybrid inference 🍯🐝] Add VAE encode (#11017) Wan Pipeline scaling fix, type hint warning, multi generator fix (#11007) Rename Lumina(2)Text2ImgPipeline -> Lumina(2)Pipeline (#10827) Quality options in `export_to_video` (#11090) Flux with Remote Encode (#11091) Don't override `torch_dtype` and don't use when `quantization_config` is set (#11039) WanI2V encode_image (#11164) Fix LatteTransformer3DModel dtype mismatch with enable_temporal_attentions (#11139) Add `latents_mean` and `latents_std` to `SDXLLongPromptWeightingPipeline` (#11034) allow models to run with a user-provided dtype map instead of a single dtype (#10301) Revert `save_model` in ModelMixin save_pretrained and use safe_serialization=False in test (#11196) [docs] `torch_dtype` map (#11194) Fix enable_sequential_cpu_offload in CogView4Pipeline (#11195) SchedulerMixin from_pretrained and ConfigMixin Self type annotation (#11192) Flux quantized with lora (#10990) AudioLDM2 Fixes (#11244) AutoModel (#11115) [docs] AutoModel (#11250) jiqing-feng (2): Enable dreambooth lora finetune example on other devices (#10602) fix autocast (#11190) kahmed10 (1): add onnxruntime-migraphx as part of check for onnxruntime in import_utils.py (#10624) kakukakujirori (1): Bug fix in LTXImageToVideoPipeline.prepare_latents() when latents is already set (#10918) kentdan3msu (1): Set self._hf_peft_config_loaded to True when LoRA is loaded using `load_lora_adapter` in PeftAdapterMixin class (#11155) lakshay sharma (1): Update import_utils.py (#10329) maxs-kan (1): Fix Flux multiple Lora loading bug (#10388) puhuk (2): Update Custom Diffusion Documentation for Multiple Concept Inference to resolve issue #10791 (#10792) Fix max_shift value in flux and related functions to 1.15 (issue #10675) (#10807) sayakpaul (1): Release: v0.33.0 sunxunle (1): chore: remove redundant words (#10609) suzukimain (2): [Docs] Added `model search` to community_projects.md (#10358) [Community] Enhanced `Model Search` (#10417) victolee0 (1): fix check_inputs func in LuminaText2ImgPipeline (#10651) wonderfan (1): chore: fix help messages in advanced diffusion examples (#10923) xieofxie (1): add provider_options in from_pretrained (#10719) yupeng1111 (1): fix wan i2v pipeline bugs (#10975) Álvaro Somoza (1): [Training] Better image interpolation in training scripts (#11206)
What does this PR do?
fix bug for torch.uint1-7 not support in torch<2.6
Before submitting
documentation guidelines, and
here are tips on formatting docstrings.
before this PR
When I execute the following command
python -c "from diffusers import AutoPipelineForText2Image"
get the following error:
The reason is that lower versions of torch do not support torch.uint1.
I added judgment on the torch version in the PR to avoid errors,and deleted redundant code.