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

Clean Up #107

Draft
wants to merge 4 commits into
base: main_perf
Choose a base branch
from
Draft

Clean Up #107

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
a286c34
update base image
micmelesse Dec 6, 2024
15a9b9c
disable navi for now
micmelesse Dec 6, 2024
9da841c
all causal seems to work on MI300
micmelesse Dec 6, 2024
e51fe14
skip MI200 causal bugs
micmelesse Dec 6, 2024
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 .github/workflows/amd_tests.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@ jobs:
id: set-matrix
run: |
if [ x"${{ github.repository }}" == x"ROCm/flash-attention" ]; then
echo '::set-output name=matrix-HIP::[["self-hosted", "gfx90a"], ["self-hosted", "gfx1100"]]'
echo '::set-output name=matrix-HIP::[["self-hosted", "gfx90a"]]'
else
echo '::set-output name=matrix-HIP::[["ubuntu-latest"]]'
fi
Expand All @@ -40,7 +40,7 @@ jobs:
matrix:
runner: ${{fromJson(needs.Runner-Preparation-AMD.outputs.matrix-HIP)}}
container:
image: rocm/pytorch:rocm6.0.2_ubuntu22.04_py3.10_pytorch_2.1.2
image: rocm/pytorch:rocm6.2.3_ubuntu22.04_py3.10_pytorch_release_2.3.0
options: --device=/dev/kfd --device=/dev/dri --security-opt seccomp=unconfined --group-add video --user root
steps:
- name: Checkout
Expand Down
1 change: 1 addition & 0 deletions .gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,7 @@ csrc/flash_attn_ck
.eggs
*.log
core.*
gpucore.*
*.csv
*.png
*.html
Expand Down
20 changes: 12 additions & 8 deletions tests/test_flash_attn_triton_amd.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -575,8 +575,8 @@ def get_dropout_fraction(
@pytest.mark.parametrize("alibi", [False])
# @pytest.mark.parametrize("local", [False, True])
@pytest.mark.parametrize("local", [False])
# @pytest.mark.parametrize("causal", [False, True])
@pytest.mark.parametrize("causal", [False])
@pytest.mark.parametrize("causal", [False, True])
# @pytest.mark.parametrize("causal", [False])
@pytest.mark.parametrize("d", [32, 40, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize('d', [32, 64, 96, 128])
Expand Down Expand Up @@ -738,8 +738,8 @@ def test_flash_attn_qkvpacked(seqlen, d, dropout_p, causal, local, alibi, determ
@pytest.mark.parametrize("alibi", [False])
# @pytest.mark.parametrize("local", [False, True])
@pytest.mark.parametrize("local", [False])
# @pytest.mark.parametrize("causal", [False, True])
@pytest.mark.parametrize('causal', [False])
@pytest.mark.parametrize("causal", [False, True])
# @pytest.mark.parametrize("causal", [False])
@pytest.mark.parametrize("d", [32, 59, 64, 80, 96, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize('d', [32])
Expand Down Expand Up @@ -898,8 +898,8 @@ def test_flash_attn_varlen_qkvpacked(
@pytest.mark.parametrize("alibi", [False])
# @pytest.mark.parametrize("local", [False, True])
@pytest.mark.parametrize("local", [False])
# @pytest.mark.parametrize("causal", [False, True])
@pytest.mark.parametrize("causal", [False])
@pytest.mark.parametrize("causal", [False, True])
# @pytest.mark.parametrize("causal", [False])
@pytest.mark.parametrize("d", [32, 40, 59, 64, 96, 111, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize('d', [32, 40, 64, 80, 96, 128, 160, 192])
Expand Down Expand Up @@ -930,6 +930,10 @@ def test_flash_attn_output(
seqlen_q, seqlen_k, d, dropout_p, causal, local, alibi, deterministic, mha_type, dtype, kvpacked, softcap
):
if USE_TRITON_ROCM:
if get_arch() == "gfx90a":
if causal == True and seqlen_q == 512 and seqlen_k == 256:
pytest.skip("This config doesnot work on MI200 Devices but works on MI300 devices.")

if softcap != 0.0:
pytest.skip("softcap not supported on AMD's Triton Backend yet")

Expand Down Expand Up @@ -1205,8 +1209,8 @@ def test_flash_attn_output(
@pytest.mark.parametrize("alibi", [False])
# @pytest.mark.parametrize("local", [False, True])
@pytest.mark.parametrize("local", [False])
# @pytest.mark.parametrize("causal", [False, True])
@pytest.mark.parametrize('causal', [False])
@pytest.mark.parametrize("causal", [False, True])
# @pytest.mark.parametrize("causal", [False])
@pytest.mark.parametrize("d", [32, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize('d', [32])
Expand Down
Loading