Chengze "Miaomiao" Li
e4cb250dc6
Merge pull request #16 from dskjal/main
...
feat: add blockswap (monolithic and detached implementation)
2026-04-19 12:44:18 +08:00
Miaomiao Li
f0833f3f20
public: fix --resolution argparse missing type=int in inference_psd.py
...
Passing --resolution <N> from CLI was kept as a string and crashed
center_square_pad_resize → smart_resize at th * tw.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-18 11:19:37 +08:00
test user
ab1933f99b
feat: optimize VRAM usage in blockswap inference pipeline
...
- enforce strict bfloat16 precision across all modules
- implement permanent CPU offloading for text encoders
- add explicit UNet block offloading before inference loop
2026-04-16 19:36:27 +09:00
Miaomiao Li
7bada45130
remove unused sys.path.append from parse_live2d
...
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-08 21:16:21 +08:00
dmMaze
4911acf64d
remove redundant parsing code from UI project loading
...
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-08 21:12:36 +08:00
dmMaze
90a77a4e53
fix default inference resolution for marigold
...
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-08 21:11:21 +08:00
Miaomiao Li
0f5ca0ea9c
public: add NF4 quantized inference for low-VRAM GPUs
2026-04-02 22:00:28 +08:00
dmMaze
92bb60494c
add --group_offload option
2026-04-02 20:45:05 +08:00
dmMaze
28a5ad21cb
cache tag embeds & unload text_encoders to save vram
2026-04-02 20:44:55 +08:00
dmMaze
1c21574924
allow specify inference resolution for depth pipeline
2026-04-02 20:44:45 +08:00
Miaomiao Li
132e54075c
public: add diffusion pipeline progress bar
...
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-02 08:28:33 +08:00
Miaomiao Li
d24c1b1435
initial release: common, inference (6 scripts + demo), annotators
2026-03-31 14:31:58 +08:00
