Support unittest for Z-image ⚡️ #12715
Conversation
…, Remove once func in pipeline.
…ryWu-code/z-image # Conflicts: # src/diffusers/models/transformers/transformer_z_image.py
…peat; Add hint for attn processor.
…ace its origin implement; Add DocString in pipeline for that.
…rd, replace its origin implement; Add DocString in pipeline for that." This reverts commit fbf26b7.
…al commit for fa3 compatibility.
… pre-encode as List of torch Tensor.
…tibility with unitest in model, mostly precision formating.
|
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. |
yiyixuxu
left a comment
yiyixuxu
left a comment
There was a problem hiding this comment.
thanks so much for the PR! I left some small suggestions:)
let me know what you think
|
Hi yiyi, this commit e277137 is ready to merged !! 🚀🚀🚀 Thanks 😊 |
yiyixuxu
left a comment
yiyixuxu
left a comment
There was a problem hiding this comment.
thanks! will merge as soon as CI passes
| # Cannot use enable_full_determinism() which sets it to True | ||
| os.environ["CUDA_LAUNCH_BLOCKING"] = "1" | ||
| os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":16:8" | ||
| torch.use_deterministic_algorithms(False) |
There was a problem hiding this comment.
I merged this but the tests are a bit flaky (it passed now and passed on my local but I think it is a bit non-deterministic)
Just making a note if it comes up in the future and we can look into it @sayakpaul @DN6
There was a problem hiding this comment.
Yeah, that would be great !!! Looing forward to that version hah, Thanks your team ~ 😊
4 participants
What does this PR do?
This PR adds unittest for Z-image Series⚡️ as discussed in #12703 (comment). Z-Image-Turbo, the distillation variant of our Z-Image, could generate 1K resolution photorealistic photo while excels in complex en/zh text rendering within 1-second in H800/H100 cards in bf16-precision.
Z-Image is a powerful and highly efficient 6B-parameter image generation model that is friendly for consumer-grade hardware, with strong capabilities in photorealistic image generation, accurate rendering of both complex Chinese and English text, and robust adherence to bilingual instructions.
The technical report and Z-Image-Turbo checkpoint will be released very soon !!!
Thanks for the support of @yiyixuxu.
Fixes # (issue)