Skip to content

Update endpoints structure #2630

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
pgmpablo157321 wants to merge 9 commits into
base: master
Choose a base branch
from
Open

Update endpoints structure #2630

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
50d4a29
Update folder structure for endpoints submissions
pgmpablo157321 Jul 14, 2026
7d8ae1a
Remove unused code
pgmpablo157321 Jul 14, 2026
5ae81f7
Revert "Remove unused code"
pgmpablo157321 Jul 14, 2026
eaa98a7
fix: skip truncation for endpoints submissions instead of targeting s…
anandhu-eng Jul 15, 2026
3e27e6f
Add compliance checking for endpoints submission (#2633)
pgmpablo157321 Jul 17, 2026
75cf495
Quick fixes for endpoints parser
pgmpablo157321 Jul 17, 2026
7e3783a
Add qwen3.6-27b benchmark to submission checker (#2634)
pgmpablo157321 Jul 17, 2026
7302435
Additional quick fixes
pgmpablo157321 Jul 17, 2026
efc2e68
Parse endpoint and performance accuracy separately
pgmpablo157321 Jul 17, 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
73 changes: 65 additions & 8 deletions tools/submission/submission_checker/checks/compliance_check.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,7 @@
from ..utils import *
import re
import os
import json


class ComplianceCheck(BaseCheck):
Expand Down Expand Up @@ -41,6 +42,7 @@ def __init__(self, log, path, config: Config,
self.submission_logs = submission_logs
self.config = config
self.name = "compliance check"
self.perf_log = self.submission_logs.performance_log
self.model = self.submission_logs.loader_data.get("benchmark", "")
self.model_mapping = self.submission_logs.loader_data.get(
"model_mapping", {})
Expand All @@ -60,14 +62,15 @@ def setup_checks(self):
Appends the per-submission validation callables to `self.checks` in
the order they should be executed by the checking framework.
"""
self.checks.append(self.dir_exists_check)
self.checks.append(self.performance_check)
self.checks.append(self.accuracy_check)
self.checks.append(self.compliance_performance_check)
self.apply_checks = set(self.checks)
# No compliance tests for endpoints for now
if self.is_endpoints:
self.apply_checks = set()
if not self.is_endpoints:
self.checks.append(self.dir_exists_check)
self.checks.append(self.performance_check)
self.checks.append(self.accuracy_check)
self.checks.append(self.compliance_performance_check)
self.apply_checks = set(self.checks)
else:
self.checks.append(self.compliance_endpoints_check)
self.apply_checks = set(self.checks)

def get_test_list(self, model):
"""Return the list of compliance tests applicable to `model`.
Expand Down Expand Up @@ -494,3 +497,57 @@ def compliance_performance_check(self):
diff)
is_valid = False
return is_valid

def compliance_endpoints_check(self):
test_dir = os.path.join(self.compliance_dir, "audit")
is_valid = True
for test in self.test_list:
if test in ENDPOINTS_COMPLIANCE_MAPPING:
test_name = ENDPOINTS_COMPLIANCE_MAPPING.get(test, test)
fname = os.path.join(test_dir, f"audit_{test_name}.json")
if not os.path.exists(fname):
self.log.error("%s is missing in %s", fname, test_dir)
is_valid = False
else:
try:
with open(fname) as f:
test_file = json.load(f)

if test_file["test"] != test_name:
self.log.error(
"%s compliance test is incorrect %s. " +
"Expected name=%s" +
"Actual name=%s",
fname,
test_dir,
test_name,
test_file["test"]
)
is_valid = False
elif not test_file["passed"]:
self.log.error(
"%s compliance test can not be loaded in %s", fname, test_dir)
is_valid = False
else:
self.log.info(
"%s compliance test %s passed for %s", fname, test_name, test_dir)

except BaseException:
self.log.error(
"%s compliance test can not be loaded in %s", fname, test_dir)
is_valid = False

if test in ["TEST09"]:
mean_output_tokens = self.perf_log["mean_output_tokens"]
if mean_output_tokens > TEST09_HIGH or mean_output_tokens < TEST09_LOW:
self.log.error("TEST 09 (Verify Output Token Length in Performance Mode)" +
" failed for %s. Mean output tokens need to be in the range" +
" [%s, %s]. Found %s",
self.compliance_dir,
TEST09_LOW,
TEST09_HIGH,
mean_output_tokens
)
is_valid = False

return is_valid
32 changes: 20 additions & 12 deletions tools/submission/submission_checker/checks/performance_check.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,10 +56,18 @@ def __init__(self, log, path, config: Config,
self.is_endpoints = self.submission_logs.loader_data.get(
"is_endpoints_submission", False)
if self.is_endpoints:
if self.scenario.lower() == "online":
self.scenario = "Server"
self.scenario = self.map_endpoints_scenario(
self.scenario, self.scenario_fixed
)
self.setup_checks()

def map_endpoints_scenario(self, scenario, scenario_fixed):
if scenario_fixed.lower() == "singlestream":
return "SingleStream"
if scenario.lower() == "online":
return "Server"
return scenario

def setup_checks(self):
"""Register individual performance-related checks.

Expand Down Expand Up @@ -492,7 +500,7 @@ def inferred_check(self):
("singlestream", "offline")
]
if (self.scenario.lower(), self.scenario_fixed.lower()
) not in list_inferred:
) not in list_inferred:
self.log.error(
"Result for scenario %s can not be inferred from %s for: %s",
self.scenario_fixed,
Expand All @@ -519,13 +527,13 @@ def get_performance_metric_check(self):
and self.mlperf_log["result_validity"] == "VALID"
):
is_valid = True
scenario = self.mlperf_log["effective_scenario"]
scenario = self.scenario
res = None
if self.is_endpoints:
if scenario.lower() == "online":
scenario = "Server"
scenario = scenario.capitalize()

res = float(self.mlperf_log[RESULT_FIELD_NEW[version][scenario]])
res = float(
self.mlperf_log[RESULT_FIELD_ENDPOINTS[version][scenario.lower()]])
else:
res = float(self.mlperf_log[RESULT_FIELD_NEW[version][scenario]])
if (
not self.is_endpoints
and version in RESULT_FIELD_BENCHMARK_OVERWRITE
Expand Down Expand Up @@ -606,12 +614,12 @@ def get_inferred_result(self, res):
res = qps_wo_loadgen_overhead

if (scenario_fixed in ["Offline"]
) and scenario in ["MultiStream"]:
) and scenario in ["MultiStream"]:
inferred = True
res = samples_per_query * S_TO_MS / (latency_mean / MS_TO_NS)

if (scenario_fixed in ["MultiStream"]
) and scenario in ["SingleStream"]:
) and scenario in ["SingleStream"]:
inferred = True
# samples_per_query does not match with the one reported in the logs
# when inferring MultiStream from SingleStream
Expand All @@ -628,6 +636,6 @@ def get_inferred_result(self, res):
else:
res = (latency_99_percentile * samples_per_query) / MS_TO_NS
if (scenario_fixed in ["Interactive"]
) and scenario not in ["Server"]:
) and scenario not in ["Server"]:
is_valid = False
return res, is_valid
87 changes: 56 additions & 31 deletions tools/submission/submission_checker/constants.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@
"gpt-oss-120b",
"wan-2.2-t2v-a14b",
"qwen3-vl-235b-a22b",
"qwen3.6-27b",
"dlrm-v3",
"yolo-95",
"yolo-99",
Expand Down Expand Up @@ -58,6 +59,7 @@
"whisper": ["Offline"],
"yolo-95": ["SingleStream", "MultiStream", "Offline"],
"yolo-99": ["SingleStream", "MultiStream", "Offline"],
"qwen3.6-27b": ["SingleStream"],
},
"optional-scenarios-edge": {},
"required-scenarios-datacenter-edge": {
Expand All @@ -79,6 +81,7 @@
"dlrm-v3": ["Offline", "Server"],
"yolo-95": ["SingleStream", "MultiStream", "Offline"],
"yolo-99": ["SingleStream", "MultiStream", "Offline"],
"qwen3.6-27b": ["SingleStream"],
},
"optional-scenarios-datacenter-edge": {
"llama2-70b-99": ["Interactive", "Server"],
Expand Down Expand Up @@ -164,6 +167,7 @@
# established
"e2e": ("E2E_ACCURACY", ""),
"e2e_vectorDB": ("E2E_ACCURACY", ""),
"qwen3.6-27b": ("mAP", 86.23 * 0.99),
},
"accuracy-upper-limit": {
"stable-diffusion-xl": (
Expand Down Expand Up @@ -203,11 +207,13 @@
"yolo-95": 64,
"yolo-99": 64,
"e2e": 824,
"e2e_vectorDB": 824
"e2e_vectorDB": 824,
"qwen3.6-27b": 995,
},
"accuracy-sample-count": {
"gpt-oss-120b": 4395,
"wan-2.2-t2v-a14b": 248,
"qwen3.6-27b": 995,
},
"dataset-size": {
"resnet": 50000,
Expand All @@ -231,6 +237,7 @@
"yolo-99": 1525,
"e2e": 824,
"e2e_vectorDB": 824,
"qwen3.6-27b": 995,
},
"model_mapping": {
"ssd-resnet34": "retinanet",
Expand Down Expand Up @@ -291,6 +298,7 @@
"yolo-99": {"SingleStream": 1024, "MultiStream": 270336, "Offline": 1},
"e2e": {"Offline": 824},
"e2e_vectorDB": {"Offline": 824},
"qwen3.6-27b": {"SingleStream": 995},
},
"models_TEST01": [
"resnet",
Expand Down Expand Up @@ -1501,6 +1509,9 @@
"compliance_accuracy.txt",
]

TEST09_LOW = 1150.38
TEST09_HIGH = 1406.02

OFFLINE_MIN_SPQ_SINCE_V4 = {
"resnet": 24576,
"retinanet": 24576,
Expand Down Expand Up @@ -1577,6 +1588,16 @@
},
}

RESULT_FIELD_ENDPOINTS = {
"v6.1": {
"offline": "result_samples_per_second",
"singlestream": "result_mean_latency_ns",
"multistream": "result_mean_latency_ns",
"server": "result_completed_samples_per_sec",
"interactive": "result_completed_samples_per_sec",
},
}

RESULT_FIELD_BENCHMARK_OVERWRITE = {
"v5.0": {
"llama2-70b-99": {
Expand Down Expand Up @@ -2030,17 +2051,19 @@
ENDPOINTS_ALLOWED_MODELS = [
"wan-2.2-t2v-a14b",
"qwen3-vl-235b-a22b",
"qwen3.6-27b",
"llama3.1-8b",
"llama3.1-8b-edge",
"gpt-oss-120b",
"deepseek-r1"
]

PERFORMANCE_ENDPOINTS_DIR = {
"v5.0": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/performance/run_1/",
"v5.1": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/performance/run_1/",
"v6.0": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/performance/run_1/",
"default": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/performance/run_1/",
ENDPOINTS_SCENARIO_DIR = {
"v5.0": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/",
"v5.1": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/",
"v6.0": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/",
"v6.1": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/",
"default": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/",
}

ACCURACY_LOG_PATH = {
Expand All @@ -2067,14 +2090,6 @@
"default": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/accuracy/mlperf_log_accuracy.json",
}

ACCURACY_ENDPOINTS_DIR = {
"v5.0": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/accuracy/",
"v5.1": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/accuracy/",
"v6.0": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/accuracy/",
"default": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/accuracy/",
}


POWER_DIR_PATH = {
"v5.0": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/performance/power",
"v5.1": "{division}/{submitter}/results/{system}/{benchmark}/{scenario}/performance/power",
Expand Down Expand Up @@ -2207,7 +2222,7 @@
"effective_sample_concatenate_permutation": "effective_sample_concatenate_permutation",
"effective_samples_per_query": "effective_samples_per_query",
"generated_query_count": "generated_query_count",
"generated_query_duration": "generated_query_duration",
"duration_ns": "generated_query_duration",
"target_qps": "effective_target_qps",
"result_scheduled_samples_per_sec": "result_scheduled_samples_per_sec",
"qps": "result_completed_samples_per_sec",
Expand All @@ -2216,30 +2231,31 @@
"latency.min": "result_min_latency_ns",
"latency.max": "result_max_latency_ns",
"latency.avg": "result_mean_latency_ns",
"latency.percentiles.50.0": "result_50.00_percentile_latency_ns",
"latency.percentiles.90.0": "result_90.00_percentile_latency_ns",
"latency.percentiles.95.0": "result_95.00_percentile_latency_ns",
"latency.percentiles.99.0": "result_99.00_percentile_latency_ns",
"latency.percentiles.50": "result_50.00_percentile_latency_ns",

@zihaok zihaok Jul 17, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

I believe the keys should still be the old format

 "percentiles": {
            "99.9": 1048838620.0,
            "99.0": 824078119.0,
            "97.0": 577244983.0,
            "95.0": 456403541.0,
            "90.0": 283834914.0,
            "80.0": 192219055.0,
            "75.0": 159208588.0,
            "50.0": 91917891.0,
            "25.0": 66838790.0,
            "10.0": 56848051.0,
            "5.0": 52593014.0,
            "1.0": 46318937.0
        },

"latency.percentiles.90": "result_90.00_percentile_latency_ns",
"latency.percentiles.95": "result_95.00_percentile_latency_ns",
"latency.percentiles.99": "result_99.00_percentile_latency_ns",
"latency.percentiles.99.9": "result_99.90_percentile_latency_ns",
"ttft.min": "result_first_token_min_latency_ns",
"ttft.max": "result_first_token_max_latency_ns",
"ttft.avg": "result_first_token_mean_latency_ns",
"ttft.percentiles.50.0": "result_first_token_50.00_percentile_latency_ns",
"ttft.percentiles.90.0": "result_first_token_90.00_percentile_latency_ns",
"ttft.percentiles.95.0": "result_first_token_95.00_percentile_latency_ns",
"ttft.percentiles.99.0": "result_first_token_99.00_percentile_latency_ns",
"ttft.percentiles.50": "result_first_token_50.00_percentile_latency_ns",
"ttft.percentiles.90": "result_first_token_90.00_percentile_latency_ns",
"ttft.percentiles.95": "result_first_token_95.00_percentile_latency_ns",
"ttft.percentiles.99": "result_first_token_99.00_percentile_latency_ns",
"ttft.percentiles.99.9": "result_first_token_99.90_percentile_latency_ns",
"tpot.percentiles.50.0": "result_time_per_output_token_50.00_percentile_ns",
"tpot.percentiles.90.0": "result_time_per_output_token_90.00_percentile_ns",
"tpot.percentiles.95.0": "result_time_per_output_token_95.00_percentile_ns",
"tpot.percentiles.99.0": "result_time_per_output_token_99.00_percentile_ns",
"tpot.percentiles.50": "result_time_per_output_token_50.00_percentile_ns",
"tpot.percentiles.90": "result_time_per_output_token_90.00_percentile_ns",
"tpot.percentiles.95": "result_time_per_output_token_95.00_percentile_ns",
"tpot.percentiles.99": "result_time_per_output_token_99.00_percentile_ns",
"tpot.percentiles.99.9": "result_time_per_output_token_99.90_percentile_ns",
"tpot.min": "result_time_per_output_token_min",
"tpot.max": "result_time_per_output_token_max",
"tpot.avg": "result_time_per_output_token_mean",
"tps": "result_completed_tokens_per_second",
"result.total": "result_query_count",
"result.failed": "num_errors",
"output_sequence_lengths.avg": "mean_output_tokens",
}


Expand All @@ -2261,9 +2277,9 @@
# Alternative JSON paths for endpoints keys that don't directly match the
# JSON structure
ENDPOINTS_JSON_ALT_PATHS = {
"result.total": "results.total",
"result.failed": "results.failed",
"qps": "results.qps",
"result.total": "n_samples_completed",
"result.failed": "n_samples_failed",
"results.qps": "qps",
"generated_query_count": "n_samples_issued",
"generated_query_duration": "duration_ns",
"test_datetime": "test_started_at",
Expand All @@ -2272,5 +2288,14 @@
}

ENDPOINTS_INFERRED_FIELDS = {
"effective_accuracy_sample_count": "result_query_count"
"effective_accuracy_sample_count": "result_query_count",
"generated_query_count": "qsl_reported_total_count",
}

ENDPOINTS_COMPLIANCE_MAPPING = {
"TEST01": "accuracy_in_performance_test",
"TEST04": "output_caching_test",
"TEST06": "consistency_output_test",
"TEST07": "accuracy_in_performance_full_test",
"TEST08": "accuracy_in_performance_dlrmv3_test",
}
Loading
Loading