Skip to content

ci: Remove cluster specific golden values #2069

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

Merged
chtruong814 merged 12 commits into from
Oct 31, 2025
Merged

ci: Remove cluster specific golden values #2069

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
d3fbbca
ci: Remove cluster specific golden values
ko3n1g Oct 31, 2025
3760ff4
chore: Update tooling for interactive jobs (#2032)
ko3n1g Oct 29, 2025
bcaccdc
chore: Update local tooling
ko3n1g Oct 31, 2025
f5b5197
ci(hotfix): Remove performance for ckpt-resume
ko3n1g Oct 31, 2025
de29be5
format
ko3n1g Oct 31, 2025
76b5089
copyright
ko3n1g Oct 31, 2025
615c7e0
rename recipes
ko3n1g Oct 31, 2025
8f5db23
Allow inference test throughput to vary by 10% (#2070)
mathemakitten Oct 31, 2025
827f0ae
ci(hotfix): Inference test pipeline
ko3n1g Oct 31, 2025
1263cc9
disable more metrics
ko3n1g Oct 31, 2025
93f906b
disable moe inf test
ko3n1g Oct 31, 2025
2e54e79
copyright
ko3n1g Oct 31, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
8 changes: 5 additions & 3 deletions tests/functional_tests/python_test_utils/test_inference_regular_pipeline.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,5 @@
# Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.

import json
import logging
import math
Expand Down Expand Up @@ -39,10 +41,10 @@ def test_inference_pipeline(golden_values_path: str, test_values_path: str) -> N
# First warmup iteration is excluded from throughput statistics.
throughput_sampled = median(output_current["throughput"][1:])

# 5 token/seconds is empirically observed to be within run variance.
# 10% is empirically observed to be within hardware variance.
assert (
throughput_sampled >= output_groundtruth["throughput"] - 5.0
), f"Throughput is slower than expected! Expected ~{output_groundtruth['throughput']} tok/s but benchmarked {output_current['throughput']} tok/s"
throughput_sampled >= 0.9 * output_groundtruth["throughput"]
), f"Throughput is slower than expected! Expected to be within 10% of ~{output_groundtruth['throughput']} tok/s but benchmarked {output_current['throughput']} tok/s"

# If throughput is significantly improved (> 20%), update golden values accordingly.
assert (
Expand Down
8 changes: 4 additions & 4 deletions tests/functional_tests/python_test_utils/test_pretraining_regular_pipeline.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,5 @@
# Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.

import logging
from typing import Dict, List, Optional

Expand All @@ -9,7 +11,7 @@
logger = logging.getLogger(__name__)

CHECK_THRESHOLDS = {
"iteration-time": [common.ApproximateTest(atol=0, rtol=0.25)],
"iteration-time": [common.ApproximateTest(atol=0, rtol=0.05)],
"mem-allocated-bytes": [common.ApproximateTest(atol=0, rtol=0.05)],
"mem-max-allocated-bytes": [common.ApproximateTest(atol=0, rtol=0.05)],
"lm loss": [common.DeterministicTest(), common.ApproximateTest(atol=0, rtol=0.05)],
Expand All @@ -32,9 +34,7 @@ def test_regular_pipeline(
model_config = yaml.safe_load(f)

checks_types = (
model_config["METRICS"]
if "METRICS" in model_config
else ["iteration-time", "lm loss", "num-zeros"]
model_config["METRICS"] if "METRICS" in model_config else ["lm loss", "num-zeros"]
)
checks = {metric: CHECK_THRESHOLDS[metric] for metric in checks_types}

Expand Down
6 changes: 3 additions & 3 deletions tests/functional_tests/python_test_utils/test_pretraining_resume_checkpoint_pipeline.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,5 @@
# Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.

import logging
from typing import Dict

Expand All @@ -20,9 +22,7 @@ def test_resume_checkpoint_pipeline(
model_config = yaml.safe_load(f)

checks_types = (
model_config["METRICS"]
if "METRICS" in model_config
else ["iteration-time", "lm loss", "num-zeros"]
model_config["METRICS"] if "METRICS" in model_config else ["lm loss", "num-zeros"]
)
checks = {
metric: test_pretraining_regular_pipeline.CHECK_THRESHOLDS[metric]
Expand Down
13 changes: 6 additions & 7 deletions tests/functional_tests/shell_test_utils/start_interactive_job.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,7 @@ DATASET_DIR=""
TIME="1:00:00"
RECIPES_DIR="tests/test_utils/recipes"
CONTAINER_MOUNTS=""
NO_GPUS_PER_TASK="FALSE"

# Declare associative array for tracking unique mounts
declare -A seen_mounts
Expand Down Expand Up @@ -53,6 +54,10 @@ while [[ $# -gt 0 ]]; do
TIME="$2"
shift 2
;;
--no-gpus-per-task)
NO_GPUS_PER_TASK="TRUE"
shift 1
;;
--help)
print_usage
exit 0
Expand All @@ -65,12 +70,6 @@ while [[ $# -gt 0 ]]; do
esac
done

# Check if yq is installed
if ! command -v yq &>/dev/null; then
echo "Error: yq is not installed. Please install it first."
exit 1
fi

# Validate required arguments
if [ -z "$PARTITION" ] || [ -z "$SLURM_ACCOUNT" ] || [ -z "$IMAGE" ] || [ -z "$DATASET_DIR" ]; then
echo "Error: Missing required arguments"
Expand All @@ -89,7 +88,7 @@ SRUN_CMD="srun \
--container-workdir=/opt/megatron-lm \
--container-mounts=$CONTAINER_MOUNTS \
--nodes=1 \
--gpus-per-task=8 \
$(if [ "$NO_GPUS_PER_TASK" = "FALSE" ]; then echo "--gpus-per-task=8"; fi) \
--time=$TIME \
--pty bash"

Expand Down
100 changes: 50 additions & 50 deletions tests/functional_tests/test_cases/bert/bert_mcore_tp1_pp2/golden_values_dev_dgx_h100.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -232,56 +232,56 @@
"end_step": 50,
"step_interval": 1,
"values": {
"1": 10.41177,
"2": 0.63219,
"3": 0.53615,
"4": 0.53244,
"5": 0.53041,
"6": 0.53364,
"7": 0.53797,
"8": 0.52807,
"9": 0.53172,
"10": 0.53116,
"11": 0.52906,
"12": 0.53113,
"13": 0.52796,
"14": 0.52974,
"15": 0.52875,
"16": 0.52005,
"17": 0.51948,
"18": 0.52008,
"19": 0.52456,
"20": 0.52593,
"21": 0.52988,
"22": 0.52281,
"23": 0.51971,
"24": 0.52235,
"25": 0.54145,
"26": 0.52876,
"27": 0.51926,
"28": 0.51381,
"29": 0.51526,
"30": 0.51632,
"31": 0.52532,
"32": 0.61496,
"33": 0.59949,
"34": 0.52069,
"35": 0.52649,
"36": 0.66485,
"37": 0.52497,
"38": 0.52464,
"39": 0.76801,
"40": 0.52465,
"41": 0.69091,
"42": 0.74369,
"43": 0.5242,
"44": 0.75825,
"45": 0.68331,
"46": 0.75831,
"47": 0.51724,
"48": 0.51305,
"49": 0.51686,
"50": 0.52176
"1": 10.44279,
"2": 0.55345,
"3": 0.53909,
"4": 0.52187,
"5": 0.52958,
"6": 0.5241,
"7": 0.5353,
"8": 0.51946,
"9": 0.52732,
"10": 0.52759,
"11": 0.51849,
"12": 0.52326,
"13": 0.52472,
"14": 0.52577,
"15": 0.51817,
"16": 0.51922,
"17": 0.51686,
"18": 0.5248,
"19": 0.51945,
"20": 0.74697,
"21": 0.51544,
"22": 0.52412,
"23": 0.66206,
"24": 0.51781,
"25": 0.52429,
"26": 0.52068,
"27": 0.62432,
"28": 0.52016,
"29": 0.52217,
"30": 0.51949,
"31": 0.69033,
"32": 0.52127,
"33": 0.52602,
"34": 0.6403,
"35": 0.51723,
"36": 0.52445,
"37": 0.51746,
"38": 0.52296,
"39": 0.52159,
"40": 0.6718,
"41": 0.58171,
"42": 0.7393,
"43": 0.54277,
"44": 0.81615,
"45": 0.52284,
"46": 0.71947,
"47": 0.52219,
"48": 0.51866,
"49": 0.51764,
"50": 0.51841
}
}
}
Loading
Loading