Skip to content

[CI] Add more ported distributed cases #2082

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

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[CI] Add more ported distributed cases #2082

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
5ab5068
add more fsdp,fsdp2 cases
zxd1997066 Sep 19, 2025
85fa6f1
add distributed UT xdist
zxd1997066 Sep 25, 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
53 changes: 39 additions & 14 deletions .github/actions/get-runner/action.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,11 @@
name: Get Runner Infos

inputs:
ut_name:
required: true
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
required: true
required: false

type: string
description: Which ut to launch

outputs:
runner_id:
value: ${{ steps.runner.outputs.runner_id }}
Expand Down Expand Up @@ -57,22 +63,41 @@ runs:
if(gpu==1 && $0~/Platform/){gpu=0}; if(gpu==1){print $0}; if($0~/Platform.*Graphics/){gpu=1}
}' |wc -l)"
cpus_per_xpu="$(echo |awk -v c="${cpu_num}" -v x="${xpu_num}" '{printf c/x}')"
pytest_extra_args="$(echo |awk -v x="${xpu_num}" -v z="${ZE_AFFINITY_MASK}" -v cx="${cpus_per_xpu}" '{
if (x > 0) {
split(z, xpu_list, ",");
for (i=0;i<x;i++) {
if (z != "") {
ze = xpu_list[i+1];
} else {
ze = i;
if [ "${{ inputs.ut_name }}" == "xpu_distributed" ];then
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there any assumptions in here? Can we detect topology directly and dynamically on the test node?
Please consider below scenarios:

  • No Xelink group, return failed
  • 1 Xelink group, launch 1 worker
  • 2 Xelink group, launch 2 workers
  • ...

pytest_extra_args="$(echo |awk -v x="${xpu_num}" -v z="${ZE_AFFINITY_MASK}" -v cx="${cpus_per_xpu}" '{
if (x > 0) {
split(z, xpu_list, ",");
for (i=0;i<x;i=i+4) {
if (z != "") {
ze = xpu_list[i+1];
} else {
ze = i;
}
printf(" --tx popen//env:ZE_AFFINITY_MASK=%d,%d,%d,%d//env:OMP_NUM_THREADS=%d//python=\"numactl -l -C %d-%d python\"",
ze,ze+1,ze+2,ze+3,4*cx,i*cx,(i+4)*cx-1);
}
printf(" --tx popen//env:ZE_AFFINITY_MASK=%d//env:OMP_NUM_THREADS=%d//python=\"numactl -l -C %d-%d python\"",
ze, cx, i*cx, (i+1)*cx-1);
}else {
printf(" -n 1 ");
}
}else {
printf(" -n 1 ");
}
}')"
}')"
else
pytest_extra_args="$(echo |awk -v x="${xpu_num}" -v z="${ZE_AFFINITY_MASK}" -v cx="${cpus_per_xpu}" '{
if (x > 0) {
split(z, xpu_list, ",");
for (i=0;i<x;i++) {
if (z != "") {
ze = xpu_list[i+1];
} else {
ze = i;
}
printf(" --tx popen//env:ZE_AFFINITY_MASK=%d//env:OMP_NUM_THREADS=%d//python=\"numactl -l -C %d-%d python\"",
ze, cx, i*cx, (i+1)*cx-1);
}
}else {
printf(" -n 1 ");
}
}')"
fi
echo "xpu_num=${xpu_num}" |tee -a ${GITHUB_OUTPUT}
echo "cpus_per_xpu=${cpus_per_xpu}" |tee -a ${GITHUB_OUTPUT}
echo "pytest_extra_args=${pytest_extra_args}" |tee -a ${GITHUB_OUTPUT}
Expand Down
6 changes: 4 additions & 2 deletions .github/workflows/_linux_ut.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,7 @@ env:
jobs:
runner:
runs-on: ${{ inputs.runner }}
name: get-runner
name: get-runner
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why we have such change?

outputs:
runner_id: ${{ steps.runner-info.outputs.runner_id }}
user_id: ${{ steps.runner-info.outputs.user_id }}
Expand All @@ -53,6 +53,8 @@ jobs:
uses: actions/checkout@v4
- name: Get runner
id: runner-info
with:
ut_name: ${{ inputs.ut }}
uses: ./.github/actions/get-runner

test-in-container:
Expand Down Expand Up @@ -104,7 +106,7 @@ jobs:
runs-on: ${{ needs.runner.outputs.runner_id }}
env:
AGENT_TOOLSDIRECTORY: /tmp/xpu-tool
PYTEST_ADDOPTS: -v --timeout 3600 --timeout_method=thread -n 1
PYTEST_ADDOPTS: -v --timeout 3600 --timeout_method=thread --dist worksteal ${{ needs.runner.outputs.pytest_extra_args }}
steps:
- name: Checkout torch-xpu-ops
uses: actions/checkout@v4
Expand Down
39 changes: 0 additions & 39 deletions test/xpu/run_distributed.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,3 @@
import os
import subprocess
import sys

Expand All @@ -9,42 +8,6 @@
res2 = 0
fail_test = []

# Get the xelink group card affinity
ret = os.system("xpu-smi topology -m 2>&1|tee topology.log")
if ret == 0:
gpu_dict = {}
with open("topology.log") as file:
lines = file.readlines()
for line in lines:
if "CPU Affinity" in line:
continue
line = line.strip()
if line.startswith("GPU "):
items = line.split(" ")
items = [x for x in items if x]
gpu_id = items[1]
i = gpu_id.split("/")[0]
affinity = ""
for j, item in enumerate(items):
if "SYS" not in item and ("XL" in item or "S" in item):
if len(affinity) == 0:
affinity = str(j - 2)
else:
affinity = affinity + "," + str(j - 2)
gpu_dict[i] = affinity

max_affinity = ""
for key, value in gpu_dict.items():
if len(value) > len(max_affinity):
max_affinity = value

os.environ["ZE_AFFINITY_MASK"] = str(max_affinity)
print(str("ZE_AFFINITY_MASK=" + os.environ.get("ZE_AFFINITY_MASK")))

else:
print("xpu-smi topology failed")
sys.exit(255)


# run python test
def run(test_command):
Expand All @@ -56,8 +19,6 @@ def run(test_command):
return result.returncode


test_command = ["python", "distributed/test_c10d_ops_xccl.py"]
res += run(test_command)
test_command = ["python", "../../../../test/distributed/pipelining/test_backward.py"]
res += run(test_command)
test_command = ["python", "../../../../test/distributed/pipelining/test_microbatch.py"]
Expand Down
85 changes: 66 additions & 19 deletions test/xpu/skip_list_dist.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,35 +1,82 @@
skip_dict = {
"../../../../test/distributed/fsdp/test_fsdp_checkpoint.py": None,
"../../../../test/distributed/fsdp/test_checkpoint_wrapper.py": None,
"../../../../test/distributed/fsdp/test_fsdp_backward_prefetch.py": None,
"../../../../test/distributed/fsdp/test_fsdp_apply.py": None,
"../../../../test/distributed/fsdp/test_fsdp_clip_grad_norm.py": (
"test_ddp_parity_xpu",
),
"../../../../test/distributed/fsdp/test_fsdp_clip_grad_norm.py": None,
"../../../../test/distributed/fsdp/test_fsdp_comm.py": None,
"../../../../test/distributed/fsdp/test_fsdp_comm_hooks.py": None,
"../../../../test/distributed/fsdp/test_fsdp_core.py": None,
"../../../../test/distributed/fsdp/test_fsdp_core.py": (
"test_transformer_no_grad_mixed_precision_True_xpu",
),
"../../../../test/distributed/fsdp/test_fsdp_exec_order.py": None,
"../../../../test/distributed/fsdp/test_fsdp_fine_tune.py": (
"test_parity_with_non_frozen_fsdp_xpu",
"test_parity_with_ddp_xpu",
"../../../../test/distributed/fsdp/test_fsdp_dtensor_state_dict.py": None,
"../../../../test/distributed/fsdp/test_fsdp_flatten_params.py": None,
"../../../../test/distributed/fsdp/test_fsdp_freezing_weights.py": None,
"../../../../test/distributed/fsdp/test_fsdp_hybrid_shard.py": None,
"../../../../test/distributed/fsdp/test_fsdp_ignored_modules.py": None,
"../../../../test/distributed/fsdp/test_fsdp_memory.py": None,
"../../../../test/distributed/fsdp/test_fsdp_meta.py": None,
"../../../../test/distributed/fsdp/test_fsdp_misc.py": None,
"../../../../test/distributed/fsdp/test_fsdp_overlap.py": None,
"../../../../test/distributed/fsdp/test_fsdp_pure_fp16.py": None,
"../../../../test/distributed/fsdp/test_fsdp_sharded_grad_scaler.py": None,
"../../../../test/distributed/fsdp/test_fsdp_state_dict.py": None,
"../../../../test/distributed/fsdp/test_fsdp_tp_integration.py": None,
"../../../../test/distributed/fsdp/test_fsdp_traversal.py": None,
"../../../../test/distributed/fsdp/test_fsdp_uneven.py": None,
"../../../../test/distributed/fsdp/test_fsdp_unshard_params.py": None,
"../../../../test/distributed/fsdp/test_fsdp_use_orig_params.py": (
"test_diff_hyperparams_sharding_strategy_str_full_shard",
"test_diff_hyperparams_sharding_strategy_str_shard_grad_op",
),
"../../../../test/distributed/fsdp/test_hsdp_dtensor_state_dict.py": None,
"../../../../test/distributed/fsdp/test_shard_utils.py": None,
"../../../../test/distributed/fsdp/test_utils.py": None,
"../../../../test/distributed/fsdp/test_wrap.py": None,
"../../../../test/distributed/fsdp/test_fsdp_fine_tune.py": None,
"../../../../test/distributed/fsdp/test_fsdp_fx.py": None,
"../../../../test/distributed/fsdp/test_fsdp_input.py": None,
"../../../../test/distributed/fsdp/test_fsdp_multiple_forward.py": None,
"../../../../test/distributed/fsdp/test_fsdp_multiple_wrapping.py": (
"test_transformer_no_grad_mixed_precision_True_xpu",
"../../../../test/distributed/fsdp/test_fsdp_multiple_wrapping.py": None,
"../../../../test/distributed/_composable/fsdp/test_fully_shard_clip_grad_norm_.py": None,
"../../../../test/distributed/_composable/fsdp/test_fully_shard_comm.py": None,
"../../../../test/distributed/_composable/fsdp/test_fully_shard_compile.py": (
"test_compiled_autograd_ctx",
"test_nested_fully_shard_backend_aot_eager",
"test_nested_fully_shard_backend_aot_eager_decomp_partition",
"test_nested_fully_shard_backend_inductor_fullgraph_True",
"test_nested_fully_shard_backend_inductor_fullgraph_True_graph_partition",
"test_simple_mlp_fullgraph_backend_aot_eager",
"test_simple_mlp_fullgraph_backend_aot_eager_decomp_partition",
"test_simple_mlp_fullgraph_backend_inductor",
"test_transformer_backend_aot_eager",
"test_transformer_backend_aot_eager_decomp_partition",
),
"../../../../test/distributed/fsdp/test_fsdp_uneven.py": None,
"../../../../test/distributed/fsdp/test_utils.py": None,
"distributed/test_c10d_xccl.py": (
# https://github.com/intel/torch-xpu-ops/issues/2046
"test_unwaited",
"../../../../test/distributed/_composable/fsdp/test_fully_shard_frozen.py": None,
"../../../../test/distributed/_composable/fsdp/test_fully_shard_mixed_precision.py": None,
"../../../../test/distributed/_composable/fsdp/test_fully_shard_state_dict.py": (
"test_cached_state_dict",
"test_dp_state_dict_cpu_offload",
),
"distributed/test_c10d_ops_xccl.py": None,
"../../../../test/distributed/fsdp/test_fsdp_misc.py": None,
"../../../../test/distributed/test_functional_api.py": (
# depends on https://github.com/pytorch/pytorch/pull/159473
"test_tracing_with_fakepg_xpu",
"../../../../test/distributed/_composable/fsdp/test_fully_shard_training.py": (
"test_explicit_prefetching",
"test_multi_forward_module",
"test_train_parity_single_group_shard_dim0",
"test_train_parity_single_group_shard_largest_dim",
"test_train_parity_shard_placement_fn_shard_largest_dim",
"test_3d_mlp_with_nd_mesh",
),
"../../../../test/distributed/_composable/test_composability/test_2d_composability.py": (
"test_tp_with_fsdp_offloading",
),
"../../../../test/distributed/_composable/test_replicate_with_compiler.py": None,
"../../../../test/distributed/_composable/test_composability/test_pp_composability.py": None,
"../../../../test/distributed/_composable/test_checkpoint.py": None,
"../../../../test/distributed/_composable/test_contract.py": None,
"distributed/test_c10d_xccl.py": None,
"distributed/test_c10d_ops_xccl.py": None,
"../../../../test/distributed/test_functional_api.py": None,
"../../../../test/distributed/_tools/test_fsdp2_mem_tracker.py": None,
"../../../../test/distributed/_tools/test_mem_tracker.py": None,
"../../../../test/distributed/_tools/test_memory_tracker.py": None,
Expand Down