rorapp/cluster-knaster
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

core: Inventar/Placement, Konfig-Validierung, CLI, HTTP-API, LiteLLM-Konfiguration, Beispiel-Skills

Browse Source 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
rorapp
2026-10-12 00:47:13 +02:00
co-authored by Claude Opus 5.5
parent f2da893f71
commit 4cbe313c62
20 changed files with 1085 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files
core/tests/test_execution.py
+10 -9
View File
@@ -41,7 +41,7 @@ def make_runner(registry, library, embedder, store, fake_chat, classifiers):
def test_weak_success_records_case(make_runner, store):
runner = make_runner(**{"rechnungen-v1": lambda m: GOOD})
runner = make_runner(**{"tier1-general/rechnungen-v1": lambda m: GOOD})
res = runner.run(RECHNUNG)
assert res.status == "ok" and res.stage == 1 and res.escalations == 0
cases = store.list_cases()
@@ -51,7 +51,7 @@ def test_weak_success_records_case(make_runner, store):
def test_schema_failure_escalates_and_creates_pair(make_runner, store):
runner = make_runner(**{
"rechnungen-v1": lambda m: json.dumps({"nummer": "4711"}), # betrag fehlt
"tier1-general/rechnungen-v1": lambda m: json.dumps({"nummer": "4711"}), # betrag fehlt
"tier0-strong": lambda m: GOOD,
})
res = runner.run(RECHNUNG)
@@ -63,7 +63,7 @@ def test_schema_failure_escalates_and_creates_pair(make_runner, store):
def test_invalid_json_signal(make_runner, store):
runner = make_runner(**{"rechnungen-v1": lambda m: "kein json", "tier0-strong": lambda m: GOOD})
runner = make_runner(**{"tier1-general/rechnungen-v1": lambda m: "kein json", "tier0-strong": lambda m: GOOD})
res = runner.run(RECHNUNG)
assert Signal.INVALID_FORMAT in res.attempts[0].execution.signals
assert res.stage == 0
@@ -74,7 +74,7 @@ def test_endpoint_error_uses_fallback_slot_without_case(make_runner, store):
def boom(m):
raise EndpointError("down")
runner = make_runner(**{"rechnungen-v1": boom})
runner = make_runner(**{"tier1-general/rechnungen-v1": boom})
# Fallback-Slot tier1-flex hat dieselbe Basis → gleicher Adapter-Name; daher Fehler nur einmal
calls = {"n": 0}
@@ -84,7 +84,8 @@ def test_endpoint_error_uses_fallback_slot_without_case(make_runner, store):
raise EndpointError("down")
return GOOD
runner.executor.chat.handlers["rechnungen-v1"] = flaky
runner.executor.chat.handlers["tier1-general/rechnungen-v1"] = flaky
runner.executor.chat.handlers["tier1-flex/rechnungen-v1"] = lambda m: GOOD
res = runner.run(RECHNUNG)
assert res.status == "ok"
assert [a.decision.slot for a in res.attempts] == ["tier1-general", "tier1-flex"]
@@ -96,7 +97,7 @@ def test_context_overflow_detected(make_runner):
def overflow(m):
raise ContextOverflowError("maximum context length")
runner = make_runner(**{"rechnungen-v1": overflow, "tier0-strong": lambda m: GOOD})
runner = make_runner(**{"tier1-general/rechnungen-v1": overflow, "tier0-strong": lambda m: GOOD})
res = runner.run(RECHNUNG)
assert res.attempts[0].execution.signals == [Signal.CONTEXT_OVERFLOW]
assert res.status == "ok"
@@ -105,7 +106,7 @@ def test_context_overflow_detected(make_runner):
def test_truncated_and_degenerate(make_runner):
loop = " ".join(["immer wieder das gleiche"] * 30)
runner = make_runner(**{
"rechnungen-v1": lambda m: ChatResult(content=loop, finish_reason="length"),
"tier1-general/rechnungen-v1": lambda m: ChatResult(content=loop, finish_reason="length"),
"tier0-strong": lambda m: GOOD,
})
res = runner.run(RECHNUNG)
@@ -114,7 +115,7 @@ def test_truncated_and_degenerate(make_runner):
def test_total_failure(make_runner):
runner = make_runner(**{"rechnungen-v1": lambda m: "x", "tier0-strong": lambda m: "y"})
runner = make_runner(**{"tier1-general/rechnungen-v1": lambda m: "x", "tier0-strong": lambda m: "y"})
res = runner.run(RECHNUNG)
assert res.status == "failed"
@@ -193,7 +194,7 @@ def test_split_is_deterministic():
def test_exports(make_runner, store):
runner = make_runner(**{"rechnungen-v1": lambda m: '{"nummer": "1"}', "tier0-strong": lambda m: GOOD})
runner = make_runner(**{"tier1-general/rechnungen-v1": lambda m: '{"nummer": "1"}', "tier0-strong": lambda m: GOOD})
for i in range(30):
runner.run(f"{RECHNUNG} Nr {i}")
train = list(export_sft(store, cluster="rechnungen", split="train"))