Adrien Toupet 89bde29cbe perf: optimize model initialization with meta device for 90% speedup ...

- Use meta device initialization to avoid unnecessary memory allocation during model creation
  - Reduces DiT and VAE initialization time by ~90% when loading to CPU
  - Explicitly delete state dicts after loading to free memory immediately
- Refactor configure_model_inference() into focused helper functions for DRY code
- Improve debug logging with timestamps and cleaner section separators for readability

2025-08-26 13:23:30 -04:00

..

common

refactor(WIP): complete memory management overhaul with proper error handling

2025-08-23 01:13:45 -04:00

core

perf: optimize model initialization with meta device for 90% speedup

2025-08-26 13:23:30 -04:00

data/image/transforms

Restored compatibility with FP8 safetensors for MPS backend; Changed the way to detect MPS device

2025-08-13 00:53:05 +08:00

interfaces

perf: optimize model initialization with meta device for 90% speedup

2025-08-26 13:23:30 -04:00

models

fix: add OOM retry mechanism and fix FP16 black frames regression

2025-08-26 00:03:02 -04:00

optimization

fix: add OOM retry mechanism and fix FP16 black frames regression

2025-08-26 00:03:02 -04:00

utils

perf: optimize model initialization with meta device for 90% speedup

2025-08-26 13:23:30 -04:00

__init__.py

refactor(WIP): Memory Management Overhaul

2025-08-22 09:20:41 -04:00