Skip to content

[FIX] torch_fused on CPU #2096

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 23, 2025
Merged

[FIX] torch_fused on CPU #2096

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 13 additions & 6 deletions gptqmodel/models/loader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -555,6 +555,7 @@ def skip(*args, **kwargs):

def build_layerwise_device_map(
model,
device,
layers: List[torch.nn.Module],
ignore_modules: List[torch.nn.Module],
num_gpus: Optional[int] = None,
Expand All @@ -581,10 +582,16 @@ def build_layerwise_device_map(
device_map: Dict[str, str] = {}
mod2name = {m: n for n, m in model.named_modules()}

if torch.cuda.is_available():
device_strs = [f"cuda:{i}" for i in range(num_gpus)]
elif hasattr(torch, "xpu") and torch.xpu.is_available():
device_strs = [f"xpu:{i}" for i in range(num_gpus)]
if device == DEVICE.CUDA:
if torch.cuda.is_available():
device_strs = [f"cuda:{i}" for i in range(num_gpus)]
else:
raise RuntimeError("CUDA is not available")
elif device == DEVICE.XPU:
if hasattr(torch, "xpu") and torch.xpu.is_available():
device_strs = [f"xpu:{i}" for i in range(num_gpus)]
else:
raise RuntimeError("XPU is not available")
else:
device_strs = ["cpu"] * num_gpus

Expand Down Expand Up @@ -653,7 +660,7 @@ def assign(mod, device_id):
else:
owner = owner.rsplit(".", 1)[0]
if owner:
device_map.setdefault(owner, fallback_device)
device_map.setdefault(owner, device_strs[fallback_device])
else:
log.info(f"Loader: unable to map param '{param_name}' to a module; skipping fallback assignment.")

Expand Down Expand Up @@ -685,7 +692,7 @@ def assign(mod, device_id):
num_gpus = torch.cuda.device_count()
elif device is DEVICE.XPU:
num_gpus = torch.xpu.device_count()
device_map = build_layerwise_device_map(model, layers, ignore_modules, num_gpus)
device_map = build_layerwise_device_map(model, device, layers, ignore_modules, num_gpus)
log.info(f"Loader: device_map = {device_map}")

load_checkpoint_in_model = True
Expand Down