Add --output-iter-metrics flag to cpu userbenchmark scripts

[murste01]

Adds a new --output-iter-metrics flag which adds per-iteration metrics to benchmark result JSON files. This allows us to do our own statistical analysis and comparison of latency/throughput.

[facebook-github-bot]

Hi @murste01!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at [email protected]. Thanks!

[murste01]

CCing @FindHao as I've seen you review similar PRs.

[FindHao]

can you add the commands to run with and w/o this config and corresponding output?

[FindHao]

default=False

[murste01]

Done.

[FindHao]

LGTM! Thank you for your contribution!
Nit:
can you add the commands to run with and w/o this config and corresponding output in this PR's description? One example is good enough. like the Test plan in this PR #2507

[facebook-github-bot]

@FindHao has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[murste01]

Of course. See below:

Without --output-iter-metrics:

Example command:

$ python3 run_benchmark.py cpu -t eval -m llama,dlrm --precision fp32 --nwarmup 10 --niter 10 -o results
                                                                                                                                                                                                                      
Running benchmark: /home/murste01/miniforge3/envs/pytorch/bin/python3 /home/murste01/git/benchmark/userbenchmark/cpu/run_config.py -m llama -d cpu -t eval --precision fp32 --metrics latencies --nwarmup 10 --niter 1
0 -o results                                                                                                                                                                                                          
Running TorchBenchModelConfig(name='llama', test='eval', device='cpu', batch_size=None, extra_args=['--precision', 'fp32'], extra_env=None, output_dir=None, skip=False) ... [Done]                                   
                                                                                                                                                                                                                      
Running benchmark: /home/murste01/miniforge3/envs/pytorch/bin/python3 /home/murste01/git/benchmark/userbenchmark/cpu/run_config.py -m dlrm -d cpu -t eval --precision fp32 --metrics latencies --nwarmup 10 --niter 10
 -o results                                                                                                                                                                                                           
Running TorchBenchModelConfig(name='dlrm', test='eval', device='cpu', batch_size=None, extra_args=['--precision', 'fp32'], extra_env=None, output_dir=None, skip=False) ... [Done]

Output tree:

$ tree
...
metrics-20250326164001.json
results/
|-- dlrm-eval
|   `-- metrics-1489.json
`-- llama-eval
    `-- metrics-1362.json
...

results/dlrm-eval/metrics-1489.json contents:

$ cat results/dlrm-eval/metrics-1489.json
{
    "name": "cpu",
    "environ": {
        "pytorch_git_version": "2236df1770800ffea5697b11b0bb0d910b2e59e1"
    },
    "metrics": {
        "latency": 19.028374499999998
    }
}

results/llama-eval/metrics-1362.json contents:

$ cat results/llama-eval/metrics-1362.json
{
    "name": "cpu",
    "environ": {
        "pytorch_git_version": "2236df1770800ffea5697b11b0bb0d910b2e59e1"
    },
    "metrics": {
        "latency": 115.062351
    }
}

metrics-20250326164001.json contents:

$ cat metrics-20250326164001.json
{
    "name": "cpu",
    "environ": {
        "pytorch_git_version": "2236df1770800ffea5697b11b0bb0d910b2e59e1"
    },
    "metrics": {
        "llama-eval_latency": 115.062351,
        "dlrm-eval_latency": 19.028374499999998
    }
}

With --output-iter-metrics:

Example command:

$ python3 run_benchmark.py cpu -t eval -m llama,dlrm --precision fp32 --nwarmup 10 --niter 10 -o results --output-iter-metrics
                                                                                                                                                                                                                      
Running benchmark: /home/murste01/miniforge3/envs/pytorch/bin/python3 /home/murste01/git/benchmark/userbenchmark/cpu/run_config.py -m llama -d cpu -t eval --precision fp32 --output-iter-metrics --metrics latencies 
--nwarmup 10 --niter 10 -o results                                                                                                                                                                                    
Running TorchBenchModelConfig(name='llama', test='eval', device='cpu', batch_size=None, extra_args=['--precision', 'fp32'], extra_env=None, output_dir=None, skip=False) ... [Done]                                   
                                                                                                                                                                                                                      
Running benchmark: /home/murste01/miniforge3/envs/pytorch/bin/python3 /home/murste01/git/benchmark/userbenchmark/cpu/run_config.py -m dlrm -d cpu -t eval --precision fp32 --output-iter-metrics --metrics latencies -
-nwarmup 10 --niter 10 -o results                                                                                                                                                                                     
Running TorchBenchModelConfig(name='dlrm', test='eval', device='cpu', batch_size=None, extra_args=['--precision', 'fp32'], extra_env=None, output_dir=None, skip=False) ... [Done]

Output Tree:

$ tree
...
metrics-20250326163144.json
results/
|-- dlrm-eval
|   `-- metrics-1088.json
`-- llama-eval
    `-- metrics-961.json
...

results/dlrm-eval/metrics-1088.json contents:

$ cat results/dlrm-eval/metrics-1088.json
{
    "name": "cpu",
    "environ": {
        "pytorch_git_version": "2236df1770800ffea5697b11b0bb0d910b2e59e1"
    },
    "metrics": {
        "latency": 19.2843725,
        "iter_latencies": [
            18.054879,
            21.023861,
            20.065568,
            20.18488,
            18.503177,
            20.390915,
            18.033985,
            21.313112,
            17.822422,
            18.1846
        ]
    }
}

results/llama-eval/metrics-991.json contents:

$ cat results/llama-eval/metrics-961.json
{
    "name": "cpu",
    "environ": {
        "pytorch_git_version": "2236df1770800ffea5697b11b0bb0d910b2e59e1"
    },
    "metrics": {
        "latency": 116.56627399999999,
        "iter_latencies": [
            124.029208,
            112.892981,
            116.560487,
            115.278424,
            101.89432,
            121.628168,
            116.572061,
            120.830798,
            132.563498,
            114.872396
        ]
    }
}

metrics-20250326163144.json contents:

$ cat metrics-20250326163144.json
{
    "name": "cpu",
    "environ": {
        "pytorch_git_version": "2236df1770800ffea5697b11b0bb0d910b2e59e1"
    },
    "metrics": {
        "llama-eval_latency": 116.56627399999999,
        "llama-eval_iter_latencies": [
            124.029208,
            112.892981,
            116.560487,
            115.278424,
            101.89432,
            121.628168,
            116.572061,
            120.830798,
            132.563498,
            114.872396
        ],
        "dlrm-eval_latency": 19.2843725,
        "dlrm-eval_iter_latencies": [
            18.054879,
            21.023861,
            20.065568,
            20.18488,
            18.503177,
            20.390915,
            18.033985,
            21.313112,
            17.822422,
            18.1846
        ]
    }
}

[facebook-github-bot]

@FindHao merged this pull request in 49c3f18.