Skip to content

chore: do not report the fake VAE "allocation" as an error#1494

Open
wbruna wants to merge 1 commit into
leejet:masterfrom
wbruna:sd_fake_vae_alloc
Open

chore: do not report the fake VAE "allocation" as an error#1494
wbruna wants to merge 1 commit into
leejet:masterfrom
wbruna:sd_fake_vae_alloc

Conversation

@wbruna
Copy link
Copy Markdown
Contributor

@wbruna wbruna commented May 15, 2026

Summary

The VAE allocation code triggered an error log for models without a VAE. I believe missing tensors due to a model error are dealt with before this point anyway, so treat the "no tensors" case as a correct path.

Additional Information

[DEBUG] ggml_extend.hpp:2746 - hidream_o1 params backend buffer size = 9485.23 MB(VRAM) (407 tensors)
[ERROR] ggml_extend.hpp:2738 - fake_vae alloc params backend buffer failed, num_tensors = 0

Checklist

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant