Skip to content

fix some dtype issue for gguf / some gpu backends#13464

Merged
yiyixuxu merged 1 commit intohuggingface:mainfrom
HsiaWinter:fix_some_dtype
Apr 14, 2026
Merged

fix some dtype issue for gguf / some gpu backends#13464
yiyixuxu merged 1 commit intohuggingface:mainfrom
HsiaWinter:fix_some_dtype

Conversation

@HsiaWinter
Copy link
Copy Markdown
Contributor

What does this PR do?

Fixes # (issue)
Fix rope dtype to float32 to support some old gpu backends and opt the time_proj dtype to support gguf

@github-actions github-actions bot added models size/S PR with diff < 50 LOC labels Apr 14, 2026
Copy link
Copy Markdown
Collaborator

@yiyixuxu yiyixuxu left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

thanks!

@HuggingFaceDocBuilderDev
Copy link
Copy Markdown

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 yiyixuxu merged commit 6a339ce into huggingface:main Apr 14, 2026
10 of 14 checks passed
yiyixuxu pushed a commit that referenced this pull request Apr 14, 2026
Document two dtype pitfalls surfaced by Ernie-Image follow-up #13464:
unconditional torch.float64 in RoPE/precompute (breaks MPS/NPU) and
reading a child module's weight dtype at runtime (breaks gguf/quant).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

models size/S PR with diff < 50 LOC

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants