Skip to content

[https://nvbugs/7043301][test] Turn off the HF weight cache in the GLM-5.2 SM103 TEP4 GSM8K test #21692

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
trtllm-agent wants to merge 4 commits into
base: main
Choose a base branch
from
+8 −1
Open

[https://nvbugs/7043301][test] Turn off the HF weight cache in the GLM-5.2 SM103 TEP4 GSM8K test #21692

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
cd3dbd3
[https://nvbugs/7043301][fix] Prefetch the checkpoint on an HF weight…
trtllm-agent Nov 5, 2026
d92ad24
[NVBUG-7043301][fix] Disable HF weight caching in GLM test
2ez4bz Nov 5, 2026
69b5ae4
[https://nvbugs/7043301][test] Say why the GLM test turns the HF weig…
trtllm-agent Nov 5, 2026
4f9a5bd
Merge remote-tracking branch 'origin/main' into repair-bot-v2/nvbug-7…
trtllm-agent Nov 6, 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
9 changes: 8 additions & 1 deletion tests/integration/defs/accuracy/modeling_v2/test_glm_moe_dsa.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -182,7 +182,7 @@ class TestModelingV2Glm52Nvfp4Sm103Tep4(LlmapiAccuracyTestHarness):

@skip_not_sm103
@pytest.mark.skip_less_device(4)
def test_gsm8k_identity_vs_mtp1_and_acceptance(self):
def test_gsm8k_identity_vs_mtp1_and_acceptance(self, monkeypatch: pytest.MonkeyPatch) -> None:
"""The identity accuracy gate, the gate on MTP not moving it, and acceptance.

Turning MTP on must not move the answers.
Expand Down Expand Up @@ -210,6 +210,13 @@ def test_gsm8k_identity_vs_mtp1_and_acceptance(self):
pass. The mtp1 leg therefore records iteration stats and checks its
acceptance length against ``ACCEPTANCE_KEY``.
"""
# Test session reuse turns the HF weight cache on for reused pools
# (tests/test_common/session_reuse.py). With it on, `auto` loads natively and
# the MTP1 leg's cache hit reads SafeTensors pages that can be dropped during
# the identity leg's GSM8K run. Turn it off so both loads use rank-striped
# read-ahead. Remove this once session reuse stops turning the cache on or the
# loader stops caching SafeTensors entries.
monkeypatch.setenv("TRTLLM_HF_WEIGHT_CACHE", "0")
task = GSM8K(self.MODEL_NAME)

# One dict for both legs: the comparison only means anything if the two
Expand Down
Loading