Adds SDXL support and CI testing, benchmarks. #271
Merged
Commits on Mar 29, 2024
-
Add precision to unet, vae and guidance scale as input to unet
Formatting Update sd_test.py Update sd_test.py Tweaks to VaeModel forward and instantiation. Fix guidance scale arg in tests. fix typo in unet_runner formatting Update VAE baseline for tests. Update VAE baseline for tests. Tweaks to vae model def and args. Fixes to encoder hidden states, remove cpu vae path
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Fix some mismatches in VAE model comparisons.
Co-authored-by: jinchen62 <[email protected]>
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Pass command line args for sdxl pytest (#487)
For running sdxl pytest on appropriate CI with different backends TODO: Run iree-tracy-capture subprocess when getting tracy profile; Build IREE from source with `-DIREE_BUILD_TRACY=ON` for CI job
-
-
-
-
Support for SDXL schedulers + example (#499)
so we don't have to carry all these branches --------- Co-authored-by: PhaneeshB <[email protected]> Co-authored-by: Avinash Sharma <[email protected]>
3 people committedMar 29, 2024 -
-
-
-
-
stateless_llama test breaks; sd_test has same issues on unet and vae: https://gist.github.com/jinchen62/11684b457c438b54a363077d0fafbe27
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Add batch count to pipeline, improve benchmarking reports, explicitly…
… use caching allocator in vmfbRunner
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Switch clip main function name and pipe through support for e2e onesh…
…ot via --compiled_pipeline option.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Set latest flags and attention spec (07f52fe)
also pipes through a classifier free option properly for turbo parity with diffusers.
-
-
-
-
-
Fix sdpa on Vulkan for SD (#557)
Co-authored-by: George Petterson <[email protected]>
-
-
-
-
-
-
-
-
-
-
-
-
-
Commits on Apr 4, 2024
Commits on Apr 8, 2024
-
fix stateless llama testing (#600)
This commit fixes the stateless llama testing. Basically, with the setup + teardown approach of pytest unit testing, all the tests were sharing the same model from hugging face. We were running the streaming llama tests before the other tests. In these tests (run_torch_llm and export_transformer_model with streaming_llm=True), we do `enable_llama_pos_shift_attention(model)`, which changes the model we are using. So, this was giving us inaccurate results by the time it came to our base vmfb_comparison test. I created this issue to track and provide more info on the error we are now seeing with torch 2.3 in `test_ test_vmfb_comparison`: #601. Also, marked it as an expected failure for now. (changed runner because we are using previous machine to repro issue for tinygrad folks which can lead to instability and system hangs)
-
-
Commits on Apr 9, 2024
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.