Compute power only in start - stop window #378
Conversation
|
MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅ |
| conversion_eff = logline.value["value"] | ||
|
|
||
| # If power stop is not found, set it to the latest power reading | ||
| if not power_stop_found: |
There was a problem hiding this comment.
is it expected that submission may not have the power-stop line? if not, maybe we should raise an error or log a warning?
There was a problem hiding this comment.
It is not expected. But it it happens, we can take the last power measurement as the stop. I updated it to log a warning
There was a problem hiding this comment.
logging/mlperf_logging/result_summarizer/result_summarizer.py
Lines 549 to 552 in 640c988
| for logline in loglines: | ||
| if logline.key == "power_measurement_start": | ||
| power_start = logline.timestamp | ||
| prev_timestamp = logline.timestamp | ||
| if logline.key == "power_reading": | ||
| agg_power += (logline.value['value'] * (logline.timestamp - prev_timestamp)) | ||
| power_start_found = True |
There was a problem hiding this comment.
if the power-start line is not found in the log, should we raise an error or log a warning? i guess it might cause an inaccurate calculation if we assumed 0 start time?
There was a problem hiding this comment.
Yes, we should rase an error. Just updated the code
There was a problem hiding this comment.
logging/mlperf_logging/result_summarizer/result_summarizer.py
Lines 546 to 548 in 640c988
pgmpablo157321
force-pushed
the
compute_power
branch
4 times, most recently
from
f990bc8
to
92c325c
Compare
pgmpablo157321
force-pushed
the
compute_power
branch
from
92c325c
to
640c988
Compare
No description provided.