File tree Expand file tree Collapse file tree 2 files changed +8
-4
lines changed
Expand file tree Collapse file tree 2 files changed +8
-4
lines changed Original file line number Diff line number Diff line change 1414import pytest
1515
1616from eval_protocol .data_loader .dynamic_data_loader import DynamicDataLoader
17- from eval_protocol .models import EvaluationRow , Message
17+ from eval_protocol .models import EvaluationRow , InputMetadata
1818from eval_protocol .pytest import evaluation_test
1919from eval_protocol .pytest .github_action_rollout_processor import GithubActionRolloutProcessor
2020
2121
2222def rows () -> List [EvaluationRow ]:
23- row = EvaluationRow (messages = [Message (role = "user" , content = "What is the capital of France?" )])
24- return [row , row , row ]
23+ return [
24+ EvaluationRow (input_metadata = InputMetadata (row_id = str (i )))
25+ for i in range (
26+ 3
27+ ) # In this example we use index to associate rows. Dataset is assumed to be accessible to the worker.
28+ ]
2529
2630
2731@pytest .mark .skipif (os .environ .get ("CI" ) == "true" , reason = "Only run this test locally (skipped in CI)" )
Original file line number Diff line number Diff line change @@ -47,7 +47,7 @@ async def test_remote_rollout_and_fetch_fireworks(row: EvaluationRow) -> Evaluat
4747 End-to-end test with Vercel production server:
4848 - Uses deployed Vercel serverless function (no manual startup needed)
4949 - trigger remote rollout via RemoteRolloutProcessor (calls init/status)
50- - fetch traces from Langfuse via Fireworks tracing proxy (uses default FireworksTracingAdapter)
50+ - fetch traces from Fireworks tracing (uses default FireworksTracingAdapter)
5151 - FAIL if no traces found or rollout_id missing
5252 """
5353 assert row .messages [0 ].content == "What is the capital of France?" , "Row should have correct message content"
You can’t perform that action at this time.
0 commit comments