Skip to content

[None][fix] revert git pin in 8f04d3a #21128

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

Closed
tburt-nv wants to merge 2 commits into from
Closed

[None][fix] revert git pin in 8f04d3a #21128

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c55d5b1
revert 8f04d3a
tburt-nv Oct 27, 2026
1959aee
Merge branch 'main' into user/tburt/revert_pin
niukuo Nov 2, 2026
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
5 changes: 0 additions & 5 deletions requirements-dev.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,11 +20,6 @@ distro==1.9.0
pyyaml>=6.0.1,<6.0.3
requests>=2.32.4,<3
boto3
# Prefix-scheduling integration tests require this exact benchmark revision.
# Owner: Simeng Liu (@SimengLiu-nv); route installation and test failures caused
# by LMBenchmark to him. Keep this revision pinned; update only when it causes
# failures, rather than tracking upstream releases on a schedule.
benchmark @ git+https://github.com/LMCache/LMBenchmark.git@c1e05a708a5a1fd04a9ec09d215edbdcccdf92cb
einops
# Reference implementation for KDA kernel parity tests.
flash-linear-attention==0.5.2
Expand Down
12 changes: 8 additions & 4 deletions tests/integration/defs/kv_cache/test_prefix_aware_scheduling.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,6 +59,9 @@
LMBENCHMARK_SCRIPT = ""
_LMBENCHMARK_PACKAGE = "benchmark"
_LMBENCHMARK_SCRIPT_IN_PACKAGE = "synthetic-multi-round-qa/multi-round-qa.py"
_LMBENCHMARK_INSTALL_SPEC = (
f"{_LMBENCHMARK_PACKAGE} @ git+https://github.com/LMCache/LMBenchmark.git@{LMBENCHMARK_SHA}"
)
CsvMetrics: TypeAlias = dict[str, int | float]
_SERVER_HOST = "0.0.0.0"
_SERVER_START_MAX_ATTEMPTS = 5
Expand Down Expand Up @@ -993,7 +996,7 @@ def _resolve_lmbenchmark_install(llm_venv: PythonVenvRunnerImpl) -> str:
dist = importlib.metadata.distribution(package)
except importlib.metadata.PackageNotFoundError:
raise SystemExit(
"LMBenchmark package is not installed. Install requirements-dev.txt before running tests. "
"LMBenchmark package is not installed after fixture setup. "
f"Expected {{script_in_package}}."
)

Expand Down Expand Up @@ -1022,9 +1025,10 @@ def _resolve_lmbenchmark_install(llm_venv: PythonVenvRunnerImpl) -> str:

@pytest.fixture(scope="module")
def ensure_lmbenchmark(llm_venv: PythonVenvRunnerImpl) -> PythonVenvRunnerImpl:
"""Verify and resolve the preinstalled LMBenchmark script used by this module."""
"""Install and resolve the pinned LMBenchmark script used by this module."""
global LMBENCHMARK_SCRIPT

llm_venv.run_cmd(["-m", "pip", "install", _LMBENCHMARK_INSTALL_SPEC], timeout=600)

@chzblych chzblych Oct 28, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

@tburt-nv Could we add retries around this install to tolerate transient network failures during Git checkout or dependency installation? The existing timeout already handles stalled installs; a small retry would help with temporary failures that exit early.

LMBENCHMARK_SCRIPT = _resolve_lmbenchmark_install(llm_venv)
return llm_venv

Expand Down Expand Up @@ -1163,8 +1167,8 @@ def test_multi_round_qa_shared_prefix(
* the CSV contains no NaN ttft values
* the server log is free of error patterns

This is a correctness stress test; request latency is reported but is
not a pass/fail threshold.
NOTE: This test is intended for post-merge / nightly CI due to its

@brnguyen2 brnguyen2 Oct 28, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

This revert brings back the "intended for post-merge / nightly CI" note, but test_multi_round_qa_shared_prefix[*] is listed in pre-merge l0_h100.yml (lines 235-242). Consider keeping the newer "correctness stress test; latency is not a pass/fail threshold" wording so the docstring doesn't contradict the test lists.

~10 min runtime per config.
"""
config_path = _write_config(tmp_path, sched_cfg)
server_log_base = str(tmp_path / "server.log")
Expand Down
Loading