[Add API] get_time_series

[yangw-dev]

Add api to get benchmark data as timeseries
currently it only support request with name: compiler_precompute,
will add general cases once we set up for torchao

this will be used as end point of benchmark regression report

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Preview	Updated (UTC)
torchci	Ready	Preview	Sep 3, 2025 8:03pm

[huydhn]

Is there a curl command that I could use to get the data from the preview API? It would make the review easier to see what the time series data looks like

[huydhn]

A curious question, do you know what is the relative size of BenchmarkTimeSeriesResponse when comparing with the raw result from ClickHouse? I want to confirm that BenchmarkTimeSeriesResponse is smaller

[yangw-dev]

it's much smaller, i think we should probably precompute those if we can.
but that is next step

[yangw-dev]

for a week's data

Tried it from postman app

url:
http://localhost:3000/api/benchmark/get_time_series

{
 "name":"compiler_precompute",
 "query_params":{
  "commits": [],
  "compilers": [],
  "arch": "h100",
  "device": "cuda",
  "dtype": "bfloat16",
  "granularity": "hour",
  "mode": "inference",
  "startTime":"2025-08-26T00:00:00",
  "stopTime": "2025-09-03T23:00:00",
  "suites": ["torchbench", "huggingface", "timm_models"],
  "workflowId": 0,
  "branches": ["main"]
 }
}

time to query

can up to 40 seconds now

size compare

• BenchmarkTimeSeriesResponse: 13.29 KB

• raw data response is more than 1.17 MB