fix: avoid f16 overflow in Z-Image quantized matmuls on ROCm#1771
Merged
Conversation
Z-Image DiT activations exceed the f16 range (up to ~1e6) in the attention qkv/out and feed_forward w2 projections. On HIP for RDNA3.x, ggml routes large-batch quantized matmuls to the hipBLAS path, which converts activations to f16, saturating them to inf; RMS_NORM then turns the inf into NaN and the decoded image comes out blank. Force f32 precision on those three projections so they take the f32 GEMM path, matching the existing Vulkan handling of w2. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Owner
|
Thank you for your contribution. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes white image generation on ROCm backend by forcing 32-bit precision.
I don't know how many other image engines are affected, but might be worth a look.