Skip to content

feat: Add support for compute capability 12.1 (Grace Blackwell) #785

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
lmxyy merged 2 commits into from
Nov 16, 2025
Merged

feat: Add support for compute capability 12.1 (Grace Blackwell) #785

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e5686dd
Add support for compute capability 12.1 (Grace Blackwell)
multippt Oct 27, 2025
bf1bfd2
Merge branch 'main' of github.com:nunchaku-tech/nunchaku into feat/sm121
lmxyy Nov 16, 2025
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
4 changes: 2 additions & 2 deletions nunchaku/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -219,7 +219,7 @@ def get_precision(
device = torch.device(device)
capability = torch.cuda.get_device_capability(0 if device.index is None else device.index)
sm = f"{capability[0]}{capability[1]}"
precision = "fp4" if sm == "120" else "int4"
precision = "fp4" if sm in ["120", "121"] else "int4"
if pretrained_model_name_or_path is not None:
if precision == "int4":
if "fp4" in str(pretrained_model_name_or_path):
Expand Down Expand Up @@ -305,7 +305,7 @@ def check_hardware_compatibility(quantization_config: dict, device: str | torch.
device = torch.device(device)
capability = torch.cuda.get_device_capability(0 if device.index is None else device.index)
sm = f"{capability[0]}{capability[1]}"
if sm == "120": # you can only use the fp4 models
if sm in ["120", "121"]: # you can only use the fp4 models
if quantization_config["weight"]["dtype"] != "fp4_e2m1_all":
raise ValueError('Please use "fp4" quantization for Blackwell GPUs. ')
elif sm in ["75", "80", "86", "89"]:
Expand Down
8 changes: 4 additions & 4 deletions setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,16 +46,16 @@ def get_sm_targets() -> list[str]:
for i in range(torch.cuda.device_count()):
capability = torch.cuda.get_device_capability(i)
sm = f"{capability[0]}{capability[1]}"
if sm == "120" and support_sm120:
sm = "120a"
assert sm in ["75", "80", "86", "89", "120a"], f"Unsupported SM {sm}"
if sm in ["120", "121"] and support_sm120:
sm = sm + "a"
assert sm in ["75", "80", "86", "89", "120a", "121a"], f"Unsupported SM {sm}"
if sm not in ret:
ret.append(sm)
else:
assert install_mode == "ALL"
ret = ["75", "80", "86", "89"]
if support_sm120:
ret.append("120a")
ret.extend(["120a", "121a"])
return ret


Expand Down