feat: create HTTP DELETE API endpoints to unregister ScoringFn and Benchmark resources in Llama Stack (#3371)

# What does this PR do?
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
This PR provides functionality for users to unregister ScoringFn and
Benchmark resources for `scoring` and `eval` APIs.

<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->
Closes #3051 

## Test Plan
<!-- Describe the tests you ran to verify your changes with result
summaries. *Provide clear instructions so the plan can be easily
re-executed.* -->
Updated integration and unit tests via CI workflow
This commit is contained in:
IAN MILLER 2025-09-15 20:43:38 +01:00 committed by GitHub
parent 01bdcce4d2
commit ab321739f2
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
13 changed files with 241 additions and 3 deletions

View file

@ -9,6 +9,7 @@ from pathlib import Path
import pandas as pd
import pytest
import requests
@pytest.fixture
@ -77,7 +78,46 @@ def test_scoring_functions_register(
assert len(list_response) > 0
assert any(x.identifier == sample_scoring_fn_id for x in list_response)
# TODO: add unregister api for scoring functions
def test_scoring_functions_unregister(
llama_stack_client,
sample_scoring_fn_id,
judge_model_id,
sample_judge_prompt_template,
):
llm_as_judge_provider = [
x
for x in llama_stack_client.providers.list()
if x.api == "scoring" and x.provider_type == "inline::llm-as-judge"
]
if len(llm_as_judge_provider) == 0:
pytest.skip("No llm-as-judge provider found, cannot test unregister")
llm_as_judge_provider_id = llm_as_judge_provider[0].provider_id
# Register first
register_scoring_function(
llama_stack_client,
llm_as_judge_provider_id,
sample_scoring_fn_id,
judge_model_id,
sample_judge_prompt_template,
)
# Ensure it is present
list_response = llama_stack_client.scoring_functions.list()
assert any(x.identifier == sample_scoring_fn_id for x in list_response)
# Unregister scoring fn
try:
base_url = llama_stack_client.base_url
except AttributeError:
pytest.skip("No server base_url available; cannot test HTTP unregister in library mode")
resp = requests.delete(f"{base_url}/v1/scoring-functions/{sample_scoring_fn_id}", timeout=30)
assert resp.status_code in (200, 204)
list_after = llama_stack_client.scoring_functions.list()
assert all(x.identifier != sample_scoring_fn_id for x in list_after)
@pytest.mark.parametrize("scoring_fn_id", ["basic::equality"])

View file

@ -105,6 +105,9 @@ class ScoringFunctionsImpl(Impl):
async def register_scoring_function(self, scoring_fn):
return scoring_fn
async def unregister_scoring_function(self, scoring_fn_id: str):
return scoring_fn_id
class BenchmarksImpl(Impl):
def __init__(self):
@ -113,6 +116,9 @@ class BenchmarksImpl(Impl):
async def register_benchmark(self, benchmark):
return benchmark
async def unregister_benchmark(self, benchmark_id: str):
return benchmark_id
class ToolGroupsImpl(Impl):
def __init__(self):
@ -330,6 +336,13 @@ async def test_scoring_functions_routing_table(cached_disk_dist_registry):
assert "test-scoring-fn" in scoring_fn_ids
assert "test-scoring-fn-2" in scoring_fn_ids
# Unregister scoring functions and verify listing
for i in range(len(scoring_functions.data)):
await table.unregister_scoring_function(scoring_functions.data[i].scoring_fn_id)
scoring_functions_list_after_deletion = await table.list_scoring_functions()
assert len(scoring_functions_list_after_deletion.data) == 0
async def test_benchmarks_routing_table(cached_disk_dist_registry):
table = BenchmarksRoutingTable({"test_provider": BenchmarksImpl()}, cached_disk_dist_registry, {})
@ -347,6 +360,15 @@ async def test_benchmarks_routing_table(cached_disk_dist_registry):
benchmark_ids = {b.identifier for b in benchmarks.data}
assert "test-benchmark" in benchmark_ids
# Unregister the benchmark and verify removal
await table.unregister_benchmark(benchmark_id="test-benchmark")
benchmarks_after = await table.list_benchmarks()
assert len(benchmarks_after.data) == 0
# Unregistering a non-existent benchmark should raise a clear error
with pytest.raises(ValueError, match="Benchmark 'dummy_benchmark' not found"):
await table.unregister_benchmark(benchmark_id="dummy_benchmark")
async def test_tool_groups_routing_table(cached_disk_dist_registry):
table = ToolGroupsRoutingTable({"test_provider": ToolGroupsImpl()}, cached_disk_dist_registry, {})

View file

@ -52,14 +52,19 @@ class TestNVIDIAEvalImpl(unittest.TestCase):
self.evaluator_post_patcher = patch(
"llama_stack.providers.remote.eval.nvidia.eval.NVIDIAEvalImpl._evaluator_post"
)
self.evaluator_delete_patcher = patch(
"llama_stack.providers.remote.eval.nvidia.eval.NVIDIAEvalImpl._evaluator_delete"
)
self.mock_evaluator_get = self.evaluator_get_patcher.start()
self.mock_evaluator_post = self.evaluator_post_patcher.start()
self.mock_evaluator_delete = self.evaluator_delete_patcher.start()
def tearDown(self):
"""Clean up after each test."""
self.evaluator_get_patcher.stop()
self.evaluator_post_patcher.stop()
self.evaluator_delete_patcher.stop()
def _assert_request_body(self, expected_json):
"""Helper method to verify request body in Evaluator POST request is correct"""
@ -115,6 +120,13 @@ class TestNVIDIAEvalImpl(unittest.TestCase):
self.mock_evaluator_post.assert_called_once()
self._assert_request_body({"namespace": benchmark.provider_id, "name": benchmark.identifier, **eval_config})
def test_unregister_benchmark(self):
# Unregister the benchmark
self.run_async(self.eval_impl.unregister_benchmark(benchmark_id=MOCK_BENCHMARK_ID))
# Verify the Evaluator API was called correctly
self.mock_evaluator_delete.assert_called_once_with(f"/v1/evaluation/configs/nvidia/{MOCK_BENCHMARK_ID}")
def test_run_eval(self):
benchmark_config = BenchmarkConfig(
eval_candidate=ModelCandidate(