Skip to content

Commit 50c114f

Browse files
arjunsuresharjunsuresh
committed
Results from GH action on NVIDIA_RTX4090x2
1 parent 6bc7c1b commit 50c114f

File tree

60 files changed

+32088
-0
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

60 files changed

+32088
-0
lines changed

‎closed/MLCommons/code/retinanet/README.md

+1
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
TBD

‎open/MLCommons/code/retinanet/README.md

+1
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
TBD

‎open/MLCommons/measurements/RTX4090x2-nvidia_original-gpu-tensorrt-vdefault-default_config/retinanet/multistream/README.md

+130
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,130 @@
1+
This experiment is generated using the [MLCommons Collective Mind automation framework (CM)](https://github.com/mlcommons/cm4mlops).
2+
3+
*Check [CM MLPerf docs](https://docs.mlcommons.org/inference) for more details.*
4+
5+
## Host platform
6+
7+
* OS version: Linux-6.2.0-39-generic-x86_64-with-glibc2.29
8+
* CPU version: x86_64
9+
* Python version: 3.8.10 (default, Sep 11 2024, 16:02:53)
10+
[GCC 9.4.0]
11+
* MLCommons CM version: 3.3.4
12+
13+
## CM Run Command
14+
15+
See [CM installation guide](https://docs.mlcommons.org/inference/install/).
16+
17+
```bash
18+
pip install -U cmind
19+
20+
cm rm cache -f
21+
22+
cm pull repo gateoverflow@cm4mlops --checkout=9aecee87a3027de8e20c63f9976584850a09ed9e
23+
24+
cm run script \
25+
--tags=app,mlperf,inference,generic,_nvidia,_retinanet,_tensorrt,_cuda,_valid,_r4.1-dev_default,_multistream \
26+
--quiet=true \
27+
--env.CM_QUIET=yes \
28+
--env.CM_MLPERF_IMPLEMENTATION=nvidia \
29+
--env.CM_MLPERF_MODEL=retinanet \
30+
--env.CM_MLPERF_RUN_STYLE=valid \
31+
--env.CM_MLPERF_SKIP_SUBMISSION_GENERATION=False \
32+
--env.CM_MLPERF_BACKEND=tensorrt \
33+
--env.CM_MLPERF_SUBMISSION_SYSTEM_TYPE=datacenter,edge \
34+
--env.CM_MLPERF_CLEAN_ALL=True \
35+
--env.CM_MLPERF_DEVICE=cuda \
36+
--env.CM_MLPERF_SUBMISSION_DIVISION=closed \
37+
--env.CM_MLPERF_USE_DOCKER=True \
38+
--env.CM_NVIDIA_GPU_NAME=rtx_4090 \
39+
--env.CM_HW_NAME=RTX4090x2 \
40+
--env.CM_RUN_MLPERF_SUBMISSION_PREPROCESSOR=yes \
41+
--env.CM_MLPERF_INFERENCE_PULL_CODE_CHANGES=yes \
42+
--env.CM_MLPERF_INFERENCE_PULL_SRC_CHANGES=yes \
43+
--env.OUTPUT_BASE_DIR=/home/arjun/gh_action_results \
44+
--env.CM_MLPERF_INFERENCE_SUBMISSION_DIR=/home/arjun/gh_action_submissions \
45+
--env.CM_MLPERF_SUBMITTER=MLCommons \
46+
--env.CM_USE_DATASET_FROM_HOST=yes \
47+
--env.CM_MLPERF_LOADGEN_ALL_SCENARIOS=yes \
48+
--env.CM_MLPERF_LOADGEN_COMPLIANCE=yes \
49+
--env.CM_MLPERF_SUBMISSION_RUN=yes \
50+
--env.CM_RUN_MLPERF_ACCURACY=on \
51+
--env.CM_RUN_SUBMISSION_CHECKER=yes \
52+
--env.CM_TAR_SUBMISSION_DIR=yes \
53+
--env.CM_MLPERF_SUBMISSION_GENERATION_STYLE=full \
54+
--env.CM_MLPERF_INFERENCE_VERSION=4.1-dev \
55+
--env.CM_RUN_MLPERF_INFERENCE_APP_DEFAULTS=r4.1-dev_default \
56+
--env.CM_MLPERF_LOADGEN_ALL_MODES=yes \
57+
--env.CM_MLPERF_INFERENCE_SOURCE_VERSION=4.1.23 \
58+
--env.CM_MLPERF_LAST_RELEASE=v4.1 \
59+
--env.CM_TMP_CURRENT_PATH=/home/arjun/actions-runner/_work/cm4mlops/cm4mlops \
60+
--env.CM_TMP_PIP_VERSION_STRING= \
61+
--env.CM_MODEL=retinanet \
62+
--env.CM_MLPERF_CLEAN_SUBMISSION_DIR=yes \
63+
--env.CM_RERUN=yes \
64+
--env.CM_MLPERF_LOADGEN_EXTRA_OPTIONS= \
65+
--env.CM_MLPERF_LOADGEN_MODE=performance \
66+
--env.CM_MLPERF_LOADGEN_SCENARIO=MultiStream \
67+
--env.CM_MLPERF_LOADGEN_SCENARIOS,=SingleStream,Offline,MultiStream,Server \
68+
--env.CM_MLPERF_LOADGEN_MODES,=performance,accuracy \
69+
--env.CM_OUTPUT_FOLDER_NAME=valid_results \
70+
--env.CM_DOCKER_REUSE_EXISTING_CONTAINER=yes \
71+
--env.CM_DOCKER_DETACHED_MODE=yes \
72+
--env.CM_MLPERF_INFERENCE_RESULTS_DIR_=/home/arjun/gh_action_results/valid_results \
73+
--env.CM_DOCKER_CONTAINER_ID=2c6007aae844 \
74+
--env.CM_MLPERF_LOADGEN_COMPLIANCE_TEST=TEST05 \
75+
--add_deps_recursive.compiler.tags=gcc \
76+
--add_deps_recursive.coco2014-original.tags=_full \
77+
--add_deps_recursive.coco2014-preprocessed.tags=_full \
78+
--add_deps_recursive.imagenet-original.tags=_full \
79+
--add_deps_recursive.imagenet-preprocessed.tags=_full \
80+
--add_deps_recursive.openimages-original.tags=_full \
81+
--add_deps_recursive.openimages-preprocessed.tags=_full \
82+
--add_deps_recursive.openorca-original.tags=_full \
83+
--add_deps_recursive.openorca-preprocessed.tags=_full \
84+
--add_deps_recursive.get-mlperf-inference-results-dir.tags=_version.r4_1-dev \
85+
--add_deps_recursive.get-mlperf-inference-submission-dir.tags=_version.r4_1-dev \
86+
--add_deps_recursive.mlperf-inference-nvidia-scratch-space.tags=_version.r4_1-dev \
87+
--adr.compiler.tags=gcc \
88+
--adr.coco2014-original.tags=_full \
89+
--adr.coco2014-preprocessed.tags=_full \
90+
--adr.imagenet-original.tags=_full \
91+
--adr.imagenet-preprocessed.tags=_full \
92+
--adr.openimages-original.tags=_full \
93+
--adr.openimages-preprocessed.tags=_full \
94+
--adr.openorca-original.tags=_full \
95+
--adr.openorca-preprocessed.tags=_full \
96+
--adr.get-mlperf-inference-results-dir.tags=_version.r4_1-dev \
97+
--adr.get-mlperf-inference-submission-dir.tags=_version.r4_1-dev \
98+
--adr.mlperf-inference-nvidia-scratch-space.tags=_version.r4_1-dev \
99+
--v=False \
100+
--print_env=False \
101+
--print_deps=False \
102+
--dump_version_info=True \
103+
--env.CM_DATASET_OPENIMAGES_PATH=/home/cmuser/CM/repos/local/cache/153f0b5852174b00/install/validation/data \
104+
--env.CM_OPENIMAGES_CALIBRATION_DATASET_PATH=/home/cmuser/CM/repos/local/cache/153f0b5852174b00/install/calibration/data \
105+
--env.CM_DATASET_OPENIMAGES_ANNOTATIONS_DIR_PATH=/home/cmuser/CM/repos/local/cache/ff5e706b363644a5 \
106+
--env.OUTPUT_BASE_DIR=/cm-mount/home/arjun/gh_action_results \
107+
--env.CM_MLPERF_INFERENCE_SUBMISSION_DIR=/cm-mount/home/arjun/gh_action_submissions \
108+
--env.MLPERF_SCRATCH_PATH=/home/cmuser/CM/repos/local/cache/e066920512fd47b7
109+
```
110+
*Note that if you want to use the [latest automation recipes](https://docs.mlcommons.org/inference) for MLPerf (CM scripts),
111+
you should simply reload gateoverflow@cm4mlops without checkout and clean CM cache as follows:*
112+
113+
```bash
114+
cm rm repo gateoverflow@cm4mlops
115+
cm pull repo gateoverflow@cm4mlops
116+
cm rm cache -f
117+
118+
```
119+
120+
## Results
121+
122+
Platform: RTX4090x2-nvidia_original-gpu-tensorrt-vdefault-default_config
123+
124+
Model Precision: int8
125+
126+
### Accuracy Results
127+
`mAP`: `37.376`, Required accuracy for closed division `>= 37.1745`
128+
129+
### Performance Results
130+
`Samples per query`: `6420049.0`

‎open/MLCommons/measurements/RTX4090x2-nvidia_original-gpu-tensorrt-vdefault-default_config/retinanet/multistream/RTX4090x2-nvidia_original-gpu-tensorrt-vdefault-default_config.json

+7
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,7 @@
1+
{
2+
"starting_weights_filename": "https://zenodo.org/record/6617981/files/resnext50_32x4d_fpn.pth",
3+
"retraining": "no",
4+
"input_data_types": "int8",
5+
"weight_data_types": "int8",
6+
"weight_transformations": "quantization, affine fusion"
7+
}

‎open/MLCommons/measurements/RTX4090x2-nvidia_original-gpu-tensorrt-vdefault-default_config/retinanet/multistream/accuracy_console.out

Whitespace-only changes.

0 commit comments

Comments
 (0)