* begin transformer conversion
* refactor
* refactor
* refactor
* refactor
* refactor
* refactor
* update
* add conversion script
* add pipeline
* make fix-copies
* remove einops
* update docs
* gradient checkpointing
* add transformer test
* update
* debug
* remove prints
* match sigmas
* add vae pt. 1
* finish CV* vae
* update
* update
* update
* update
* update
* update
* make fix-copies
* update
* make fix-copies
* fix
* update
* update
* make fix-copies
* update
* update tests
* handle device and dtype for safety checker; required in latest diffusers
* remove enable_gqa and use repeat_interleave instead
* enforce safety checker; use dummy checker in fast tests
* add review suggestion for ONNX export
Co-Authored-By: Asfiya Baig <asfiyab@nvidia.com>
* fix safety_checker issues when not passed explicitly
We could either do what's done in this commit, or update the Cosmos examples to explicitly pass the safety checker
* use cosmos guardrail package
* auto format docs
* update conversion script to support 14B models
* update name CosmosPipeline -> CosmosTextToWorldPipeline
* update docs
* fix docs
* fix group offload test failing for vae
---------
Co-authored-by: Asfiya Baig <asfiyab@nvidia.com>
* update
* update
* make style
* remove dynamo disable
* add coauthor
Co-Authored-By: Dhruv Nair <dhruv.nair@gmail.com>
* update
* update
* update
* update mixin
* add some basic tests
* update
* update
* non_blocking
* improvements
* update
* norm.* -> norm
* apply suggestions from review
* add example
* update hook implementation to the latest changes from pyramid attention broadcast
* deinitialize should raise an error
* update doc page
* Apply suggestions from code review
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>
* update docs
* update
* refactor
* fix _always_upcast_modules for asym ae and vq_model
* fix lumina embedding forward to not depend on weight dtype
* refactor tests
* add simple lora inference tests
* _always_upcast_modules -> _precision_sensitive_module_patterns
* remove todo comments about review; revert changes to self.dtype in unets because .dtype on ModelMixin should be able to handle fp8 weight case
* check layer dtypes in lora test
* fix UNet1DModelTests::test_layerwise_upcasting_inference
* _precision_sensitive_module_patterns -> _skip_layerwise_casting_patterns based on feedback
* skip test in NCSNppModelTests
* skip tests for AutoencoderTinyTests
* skip tests for AutoencoderOobleckTests
* skip tests for UNet1DModelTests - unsupported pytorch operations
* layerwise_upcasting -> layerwise_casting
* skip tests for UNetRLModelTests; needs next pytorch release for currently unimplemented operation support
* add layerwise fp8 pipeline test
* use xfail
* Apply suggestions from code review
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>
* add assertion with fp32 comparison; add tolerance to fp8-fp32 vs fp32-fp32 comparison (required for a few models' test to pass)
* add note about memory consumption on tesla CI runner for failing test
---------
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>
* copy transformer
* copy vae
* copy pipeline
* make fix-copies
* refactor; make original code work with diffusers; test latents for comparison generated with this commit
* move rope into pipeline; remove flash attention; refactor
* begin conversion script
* make style
* refactor attention
* refactor
* refactor final layer
* their mlp -> our feedforward
* make style
* add docs
* refactor layer names
* refactor modulation
* cleanup
* refactor norms
* refactor activations
* refactor single blocks attention
* refactor attention processor
* make style
* cleanup a bit
* refactor double transformer block attention
* update mochi attn proc
* use diffusers attention implementation in all modules; checkpoint for all values matching original
* remove helper functions in vae
* refactor upsample
* refactor causal conv
* refactor resnet
* refactor
* refactor
* refactor
* grad checkpointing
* autoencoder test
* fix scaling factor
* refactor clip
* refactor llama text encoding
* add coauthor
Co-Authored-By: "Gregory D. Hunkins" <greg@ollano.com>
* refactor rope; diff: 0.14990234375; reason and fix: create rope grid on cpu and move to device
Note: The following line diverges from original behaviour. We create the grid on the device, whereas
original implementation creates it on CPU and then moves it to device. This results in numerical
differences in layerwise debugging outputs, but visually it is the same.
* use diffusers timesteps embedding; diff: 0.10205078125
* rename
* convert
* update
* add tests for transformer
* add pipeline tests; text encoder 2 is not optional
* fix attention implementation for torch
* add example
* update docs
* update docs
* apply suggestions from review
* refactor vae
* update
* Apply suggestions from code review
Co-authored-by: hlky <hlky@hlky.ac>
* Update src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py
Co-authored-by: hlky <hlky@hlky.ac>
* Update src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py
Co-authored-by: hlky <hlky@hlky.ac>
* make fix-copies
* update
---------
Co-authored-by: "Gregory D. Hunkins" <greg@ollano.com>
Co-authored-by: hlky <hlky@hlky.ac>
* reduce block sizes for unet1d.
* reduce blocks for unet_2d.
* reduce block size for unet_motion
* increase channels.
* correctly increase channels.
* reduce number of layers in unet2dconditionmodel tests.
* reduce block sizes for unet2dconditionmodel tests
* reduce block sizes for unet3dconditionmodel.
* fix: test_feed_forward_chunking
* fix: test_forward_with_norm_groups
* skip spatiotemporal tests on MPS.
* reduce block size in AutoencoderKL.
* reduce block sizes for vqmodel.
* further reduce block size.
* make style.
* Empty-Commit
* reduce sizes for ConsistencyDecoderVAETests
* further reduction.
* further block reductions in AutoencoderKL and AssymetricAutoencoderKL.
* massively reduce the block size in unet2dcontionmodel.
* reduce sizes for unet3d
* fix tests in unet3d.
* reduce blocks further in motion unet.
* fix: output shape
* add attention_head_dim to the test configuration.
* remove unexpected keyword arg
* up a bit.
* groups.
* up again
* fix
