Skip to content

Commit c6aa83f

Browse files
committed
Use deferred loader for single device load
1 parent 6fc153d commit c6aa83f

File tree

1 file changed

+5
-0
lines changed

1 file changed

+5
-0
lines changed

exllamav3/models/model.py

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -92,7 +92,12 @@ def unload(self):
9292
def _load_single(self, progressbar: bool, device: torch.device):
9393
with ProgressBar(f"Loading" if progressbar else None, len(self.modules)) as progress:
9494
for idx, module in enumerate(self.modules):
95+
defer = module.can_defer_load()
96+
if defer:
97+
self.config.stc.begin_deferred_load()
9598
module.load(torch.device("cpu") if module.caps.get("prefer_cpu") else device)
99+
if defer:
100+
self.config.stc.end_deferred_load()
96101
progress.update(idx + 1)
97102

98103

0 commit comments

Comments
 (0)