implement html output, fix log scaling in charts #218
Conversation
|
📦 Build Artifacts Available |
|
📦 Build Artifacts Available |
|
🎉 Live Preview: Click here to view the live version |
|
📦 Build Artifacts Available |
There was a problem hiding this comment.
I didn't notice any major red flags from the code review.
I took the time to run a new benchmark using this branch, since the output format changed from the addition of the 50th percentile. I merged in my from-file changes for testing, and converted the JSON output to HTML, and I encountered two things.
First, I got these warnings:
Saving benchmarks report...
/Users/joconnel/Documents/projects/ai/guidellm/.venv/lib/python3.13/site-packages/pydantic/main.py:463: UserWarning: Pydantic serializer warnings:
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p50', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p90', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p95', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p99', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p50', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p90', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p95', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p99', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p50', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p90', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p95', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p99', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p50', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p90', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p95', input_type=str])
PydanticSerializationUnexpectedValue(Expected `float` - serialized value may not be as expected [input_value='p99', input_type=str])
return self.__pydantic_serializer__.to_python(
Benchmarks report saved to benchmarks_html.html
Second, the HTML file displays nearly blank. I'll upload my benchmarks file for comparison. It's possible that I did something wrong. The HTML cannot be included in this comment, so here is a picture of the output on Firefox:
benchmarks_html.json
To clarify all of the UI changes here, there was a previous version of the UI that worked with an older version of guidellm and the code used to generate the html report provided the UI with a different data format. Most of the UI changes here are needed to get things up to date with the current version of guidellm and how the html reporting logic generates data.
There are a few specific changes in these files to get logarithmic scaling working which I found was broken after making these changes, in hindsight I would have split these out into another PR:
src/ui/lib/components/Charts/DashedLine/DashedLine.component.tsx
src/ui/lib/components/Charts/DashedLine/helpers.ts
src/ui/lib/components/Charts/MetricLine/MetricLine.component.tsx
tests/ui/unit/components/Charts/DashedLine/helpers.test.ts
There were a few other tiny bug fixes as well.