diffusers
Reduce peak VRAM by releasing large attention tensors (as soon as they're unnecessary)
#3463
Merged
Overview
Commits
1
Changes
View On GitHub

Reduce peak VRAM by releasing large attention tensors (as soon as they're unnecessary) #3463

patrickvonplaten merged 1 commit into huggingface:main from cmdr2:attn_tensor_free
cmdr2
cmdr2 Release large tensors in attention (as soon as they're no longer requ…
5b68fa1b
cmdr2 cmdr2 changed the title Release large attention tensors (as soon as they're unnecessary). Reduces peak VRAM by nearly 2 GB for 1024x1024 in pre-torch 2.0 (even after slicing). More savings with larger images. Reduces peak VRAM by releasing large attention tensors (as soon as they're unnecessary) 2 years ago
cmdr2 cmdr2 changed the title Reduces peak VRAM by releasing large attention tensors (as soon as they're unnecessary) Reduce peak VRAM by releasing large attention tensors (as soon as they're unnecessary) 2 years ago
HuggingFaceDocBuilderDev
pcuenca
pcuenca approved these changes on 2023-05-17
patrickvonplaten
patrickvonplaten approved these changes on 2023-05-17
patrickvonplaten patrickvonplaten merged bd78f63a into main 2 years ago
cmdr2

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Terms Of Use Privacy Policy Refund Policy Impressum