Skip to content
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

Fix parsing v100s in get_available_flops #18952

Merged
merged 5 commits into from
Nov 6, 2023
Merged

Fix parsing v100s in get_available_flops #18952

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
f775317
fix
awaelchli Nov 5, 2023
48c4a7e
chlog
awaelchli Nov 5, 2023
0361ad1
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Nov 5, 2023
5af3c1c
Apply suggestions from code review
awaelchli Nov 6, 2023
d8f1449
V100 sxm2
carmocca Nov 6, 2023
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
3 changes: 3 additions & 0 deletions src/lightning/fabric/CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
- Fixed layer conversion under `Fabric.init_module()` context manager when using the `BitsandbytesPrecision` plugin ([#18914](https://github.com/Lightning-AI/lightning/pull/18914))


- Fixed parsing of v100s GPUs in `get_available_flops` ([#18952](https://github.com/Lightning-AI/lightning/pull/18952))


## [2.1.0] - 2023-10-11

### Added
Expand Down
2 changes: 2 additions & 0 deletions src/lightning/fabric/utilities/throughput.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -382,6 +382,8 @@ def get_available_flops(device: torch.device, dtype: Union[torch.dtype, str]) ->
chip = "v100-sxm"
elif "v100-pcie" in chip:
chip = "v100-pcie"
elif "v100s-pcie" in chip:
chip = "v100s-pcie"
elif "t4" in chip:
chip = "t4"
elif "quadro rtx 5000" in chip:
Expand Down
26 changes: 25 additions & 1 deletion tests/tests_fabric/utilities/test_throughput.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@ def test_measure_flops():
assert fwd_flops < fwd_and_bwd_flops


def test_available_flops(xla_available):
def test_get_available_flops(xla_available):
with mock.patch("torch.cuda.get_device_name", return_value="NVIDIA H100 PCIe"):
flops = get_available_flops(torch.device("cuda"), torch.bfloat16)
assert flops == 1.513e15 / 2
Expand Down Expand Up @@ -61,6 +61,30 @@ def test_available_flops(xla_available):
tpu.reset_mock()


@pytest.mark.parametrize(
"device_name",
[
# TODO: We need to represent the real names here
"h100-hbm3",
"NVIDIA H100 PCIe",
"h100-hbm2e",
"NVIDIA A100 80GB PCIe",
"NVIDIA A100-SXM4-40GB",
"NVIDIA A10G",
"Tesla V100-SXm2-32GB",
"Tesla V100-PCIE-32GB",
"Tesla V100S-PCIE-32GB",
"Tesla T4",
"Quadro RTX 5000 with Max-Q Design",
],
)
@mock.patch("lightning.fabric.accelerators.cuda._is_ampere_or_later", return_value=False)
def test_get_available_flops_cuda_mapping_exists(_, device_name):
"""Tests `get_available_flops` against known device names."""
with mock.patch("lightning.fabric.utilities.throughput.torch.cuda.get_device_name", return_value=device_name):
assert get_available_flops(device=torch.device("cuda"), dtype=torch.float32) is not None


def test_throughput():
# required args only
throughput = Throughput()
Expand Down